diff --git a/19.9m-g008-ai-r17.5/README.md b/19.9m-g008-ai-r17.5/README.md index c37e4ce9b9adb42f2159deee835cf1086e8cfc82..b5d2309490a6dca7e092c7de7214565a21f69228 100644 --- a/19.9m-g008-ai-r17.5/README.md +++ b/19.9m-g008-ai-r17.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g008-ai-r2.13/README.md b/19.9m-g008-ai-r2.13/README.md index bdc78c7ebb59b35f5c1a515bd85d2e1f0c87157c..d9d512e830656825172011d1300410db6fa4670b 100644 --- a/19.9m-g008-ai-r2.13/README.md +++ b/19.9m-g008-ai-r2.13/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g008-ai-r4.32/README.md b/19.9m-g008-ai-r4.32/README.md index ab099f6133e0aaf24a2b302f59b790f23b6f1439..65eae9a3a68dffbfcb6be123c9b82463650694a7 100644 --- a/19.9m-g008-ai-r4.32/README.md +++ b/19.9m-g008-ai-r4.32/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g008-control/README.md b/19.9m-g008-control/README.md index ba9b228f7b9ca3453614a4e3271083af6d3bc8e4..c4f63d1a256723017e09f511eb843c84aab4e215 100644 --- a/19.9m-g008-control/README.md +++ b/19.9m-g008-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g009-ai-r17.5/README.md b/19.9m-g009-ai-r17.5/README.md index 26a472359e44189fe1d20adaf25ec8c601cbcd40..044b5f6cb2b77441d3b35cdf42e02f1dd3ad6dcd 100644 --- a/19.9m-g009-ai-r17.5/README.md +++ b/19.9m-g009-ai-r17.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g009-ai-r2.13/README.md b/19.9m-g009-ai-r2.13/README.md index ab6dc0a6f7df7876d4c0193c140cfe74a1b7b9b7..fd05380bd499b166834010792823ecd52267b067 100644 --- a/19.9m-g009-ai-r2.13/README.md +++ b/19.9m-g009-ai-r2.13/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g009-ai-r4.32/README.md b/19.9m-g009-ai-r4.32/README.md index 2c7ac75deda43934878609beadd83134cd575312..9982dad8d2c3fe2aec119aa8693e5f144cd0f1d8 100644 --- a/19.9m-g009-ai-r4.32/README.md +++ b/19.9m-g009-ai-r4.32/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g009-control/README.md b/19.9m-g009-control/README.md index f24d4985023665535ff8d5a5fe9e6c96b0ee659a..48396f079c28466d100f36c63edc3903b799fa87 100644 --- a/19.9m-g009-control/README.md +++ b/19.9m-g009-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r0.025/README.md b/19.9m-g010-ai-r0.025/README.md index 5206f71f13bd7ffb3b7a385a10ffef0b7aa2cd2b..3762b5538e471f8efd86c4e205b301ab252516cc 100644 --- a/19.9m-g010-ai-r0.025/README.md +++ b/19.9m-g010-ai-r0.025/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r0.05/README.md b/19.9m-g010-ai-r0.05/README.md index db40d2eee2642e9bfaf52cc1132df3c533ac4be4..ca00f507d84ec20e04b5c747a2b085e9ce862686 100644 --- a/19.9m-g010-ai-r0.05/README.md +++ b/19.9m-g010-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r0.1/README.md b/19.9m-g010-ai-r0.1/README.md index 5f89140df511bbc7b076638e5c394e8c90aa3d0d..82557570017038b50427240b1477c39ab940ea24 100644 --- a/19.9m-g010-ai-r0.1/README.md +++ b/19.9m-g010-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r0.25/README.md b/19.9m-g010-ai-r0.25/README.md index 0f355ad5c9e29adff552816c8c96772cac543aae..b0b41c78f7d250bf82662e1ea573146bb98f5963 100644 --- a/19.9m-g010-ai-r0.25/README.md +++ b/19.9m-g010-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r0.5/README.md b/19.9m-g010-ai-r0.5/README.md index 3b9b6223a703522426dd8cd3a8b337d6e6eb801b..9f8076e856dce821a50aefd10947df7be9574b46 100644 --- a/19.9m-g010-ai-r0.5/README.md +++ b/19.9m-g010-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r1/README.md b/19.9m-g010-ai-r1/README.md index 5fd42d0586299af29c3e795da583bd8c0780f8a6..473e44c58d6277824ab5d9bf2ab844d6447fd86d 100644 --- a/19.9m-g010-ai-r1/README.md +++ b/19.9m-g010-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r16/README.md b/19.9m-g010-ai-r16/README.md index af471d63c289d29a00d6a09f1052297df0048d91..b66023c2005d8f03d3d3d67df11074aedf61b30a 100644 --- a/19.9m-g010-ai-r16/README.md +++ b/19.9m-g010-ai-r16/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r2/README.md b/19.9m-g010-ai-r2/README.md index c846d7cebebf028698dd72bc9a50efbc86aa37ef..7ef1b386dfa19c1b0e2d50c152ed4e98726a2937 100644 --- a/19.9m-g010-ai-r2/README.md +++ b/19.9m-g010-ai-r2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r32/README.md b/19.9m-g010-ai-r32/README.md index 2ae653fb3d793948eff403098c4ad653fe0752bc..ab9bd3f811e1b560a3ae9c6bd4a9c216c8922c31 100644 --- a/19.9m-g010-ai-r32/README.md +++ b/19.9m-g010-ai-r32/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r4/README.md b/19.9m-g010-ai-r4/README.md index 03616b982aa81e21b9cc4f134807f424e7ae494d..e16ee0a8cac1a89af71cd5dfb89569852f17c68e 100644 --- a/19.9m-g010-ai-r4/README.md +++ b/19.9m-g010-ai-r4/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-ai-r8/README.md b/19.9m-g010-ai-r8/README.md index 4ea040306b2dc91f2dbb5421b4393a728e19e430..ad4a433db3b52e2e67252482c8b357a1a601a876 100644 --- a/19.9m-g010-ai-r8/README.md +++ b/19.9m-g010-ai-r8/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-control/README.md b/19.9m-g010-control/README.md index da6286399fd1aab7dab31588ed4fe672dd599827..327b1cf7ba2e3ada4cef3b3ef864d92b483c522a 100644 --- a/19.9m-g010-control/README.md +++ b/19.9m-g010-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-human-r0.025/README.md b/19.9m-g010-human-r0.025/README.md index 3b6a73c73854c1987e6d9fd6281972c502f27872..d3cfc94097548bc6304c2d841126834a56eac515 100644 --- a/19.9m-g010-human-r0.025/README.md +++ b/19.9m-g010-human-r0.025/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-human-r0.05/README.md b/19.9m-g010-human-r0.05/README.md index 31e15be3931a3b59f4b747199ef14b05d0deba90..b34d61126eb992c8545bf76806531e308265c4f0 100644 --- a/19.9m-g010-human-r0.05/README.md +++ b/19.9m-g010-human-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-human-r2/README.md b/19.9m-g010-human-r2/README.md index 90642f3e14d8fa16f06473483a3ada42a417cd48..916ba5f789e9a6d16e684fdd78d4fd7c6d174736 100644 --- a/19.9m-g010-human-r2/README.md +++ b/19.9m-g010-human-r2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-human-r4/README.md b/19.9m-g010-human-r4/README.md index f7141c3aca10985864a19f75b16b9df73ba1e144..188fdfcee3f6d4238fdfdaf8d1d09105403f7cd2 100644 --- a/19.9m-g010-human-r4/README.md +++ b/19.9m-g010-human-r4/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-repeat-r1/README.md b/19.9m-g010-repeat-r1/README.md index 16fd4430d97a2471aac60c04245ff6f138cb39f8..cd580224d42caa667c7aa63fe66a4a6163384905 100644 --- a/19.9m-g010-repeat-r1/README.md +++ b/19.9m-g010-repeat-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -71,7 +71,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-repeat-r4/README.md b/19.9m-g010-repeat-r4/README.md index f6c038b0b9fca8201f85ddeb3aa084a214e239a3..2756cea273ee645206bee9525267667acb316073 100644 --- a/19.9m-g010-repeat-r4/README.md +++ b/19.9m-g010-repeat-r4/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -71,7 +71,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g010-repeat-r8/README.md b/19.9m-g010-repeat-r8/README.md index 404c2f106927494f7fceae40c43246051137ec0b..53f5d44ad44d330c79d9bcd1a4fbb6d806271336 100644 --- a/19.9m-g010-repeat-r8/README.md +++ b/19.9m-g010-repeat-r8/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -71,7 +71,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-ai-r0.05/README.md b/19.9m-g011-ai-r0.05/README.md index 5db78e647b4aaf7ed24178fd7dcafb4b0d12be72..2e45a930518d37c1b91ab4c25baabc2ec770a847 100644 --- a/19.9m-g011-ai-r0.05/README.md +++ b/19.9m-g011-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-ai-r0.1/README.md b/19.9m-g011-ai-r0.1/README.md index 17b66a8b8473845b34fde0552f3b7fe0654d3edb..24f40d53975fee122dc997a17819cc722784e881 100644 --- a/19.9m-g011-ai-r0.1/README.md +++ b/19.9m-g011-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-ai-r0.25/README.md b/19.9m-g011-ai-r0.25/README.md index 9572d7d9985fb68584aa321bc58d45b0a411d370..15818efe1ace9b6467b1d2c2c417171b5d330efa 100644 --- a/19.9m-g011-ai-r0.25/README.md +++ b/19.9m-g011-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-ai-r0.5/README.md b/19.9m-g011-ai-r0.5/README.md index 49d3767127a3af6efd8441db06948eb16581e524..7666dcdcc1c63f6e023a0e37885d6753a7eefd96 100644 --- a/19.9m-g011-ai-r0.5/README.md +++ b/19.9m-g011-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-ai-r1/README.md b/19.9m-g011-ai-r1/README.md index 78f1051cfb4529b1e84ddec7f8d0aa1e85640043..65e179291e88272d97edbd1f258dc74db915385c 100644 --- a/19.9m-g011-ai-r1/README.md +++ b/19.9m-g011-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g011-control/README.md b/19.9m-g011-control/README.md index 5f4c3d2fde99ebb7bc2a0132aad9aa2adce5134d..986bdafc50493b097577fbbdce17c546eefd9e78 100644 --- a/19.9m-g011-control/README.md +++ b/19.9m-g011-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-ai-r0.05/README.md b/19.9m-g012-ai-r0.05/README.md index 5f4309ef047d4a6805861b490944dda6afe199ff..2c253a3ebdcecd72316b30d902c51a440e035864 100644 --- a/19.9m-g012-ai-r0.05/README.md +++ b/19.9m-g012-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-ai-r0.1/README.md b/19.9m-g012-ai-r0.1/README.md index e9e482af4b8dd8beb51f57e3016c4d5d9575d3d7..d5272013aaf9d4f1c1d13e3f752d6f8f02bbf2e1 100644 --- a/19.9m-g012-ai-r0.1/README.md +++ b/19.9m-g012-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-ai-r0.25/README.md b/19.9m-g012-ai-r0.25/README.md index 279d41240b030057af746d5219ad3704141a40ec..3144b09ab0d60c5e8575502dd64bf78bd0e0ef0a 100644 --- a/19.9m-g012-ai-r0.25/README.md +++ b/19.9m-g012-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-ai-r0.5/README.md b/19.9m-g012-ai-r0.5/README.md index 02dee32efd6def9470332c3db4932311f49b1841..4129a69658c5771befba62d633b84b0f7bb524b4 100644 --- a/19.9m-g012-ai-r0.5/README.md +++ b/19.9m-g012-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-ai-r1/README.md b/19.9m-g012-ai-r1/README.md index 889a7af40f5eb7ea1a4cb76cb7d048ebfeb03e2a..3fb5aa100f1298c0c66541e1d7edf070a772f75e 100644 --- a/19.9m-g012-ai-r1/README.md +++ b/19.9m-g012-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g012-control/README.md b/19.9m-g012-control/README.md index 6fd4d49b7d99ffd0c42c5ec35b83364181560f53..7eaec90961e0b89710cceed3727668a258cc5f4c 100644 --- a/19.9m-g012-control/README.md +++ b/19.9m-g012-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-ai-r17.2/README.md b/19.9m-g013-ai-r17.2/README.md index ddad34d1560cbb2a2a44a8ac1d64cd207a2014b7..b6db11d95dc8b48fbb2e72307292a2fef14d6c2d 100644 --- a/19.9m-g013-ai-r17.2/README.md +++ b/19.9m-g013-ai-r17.2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-ai-r2.1/README.md b/19.9m-g013-ai-r2.1/README.md index 1e042564920020ffbc688cc4e13dbfd9d50c780f..c62e96ea97d9ac7926e97376e5569425981f7550 100644 --- a/19.9m-g013-ai-r2.1/README.md +++ b/19.9m-g013-ai-r2.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-ai-r4.25/README.md b/19.9m-g013-ai-r4.25/README.md index 1bd2e813f7d101e1ed5a6c1ccd8f8d50eaff097d..d117ddb673c5ad0e3c698773883405ea3e95c4ab 100644 --- a/19.9m-g013-ai-r4.25/README.md +++ b/19.9m-g013-ai-r4.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-ai-r8.57/README.md b/19.9m-g013-ai-r8.57/README.md index 81f40c0a10cc76ad38bdca002ad4abce851b74d7..a0806a334703d7bfd930a3db7176bc3e44f64f11 100644 --- a/19.9m-g013-ai-r8.57/README.md +++ b/19.9m-g013-ai-r8.57/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-control/README.md b/19.9m-g013-control/README.md index bdb4b431f877bffeb04e5d9b5248ae9b884e3c12..4dbba504d6e84b856cec11b6e0ab11becead2ce1 100644 --- a/19.9m-g013-control/README.md +++ b/19.9m-g013-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g013-human-r2/README.md b/19.9m-g013-human-r2/README.md index df7a63650689ee5152fa0001f92d563cb980540e..f791316fc3bafe0bb379654ff6bc7ca05b235f94 100644 --- a/19.9m-g013-human-r2/README.md +++ b/19.9m-g013-human-r2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g014-ai-r0.25/README.md b/19.9m-g014-ai-r0.25/README.md index 484e5d5fe5a1e9d392a8790c6d0817c90817da53..b7a9de32e6560251fa6f87fd8fd9cf65e2dbb98e 100644 --- a/19.9m-g014-ai-r0.25/README.md +++ b/19.9m-g014-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g014-ai-r1/README.md b/19.9m-g014-ai-r1/README.md index 5b440e0c4acdbc9bd91fab729ac2fe322abd9d26..aad623f6414668529e4f3a431a6e569c63d67700 100644 --- a/19.9m-g014-ai-r1/README.md +++ b/19.9m-g014-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g014-control/README.md b/19.9m-g014-control/README.md index 81903a81fbb16b1f4f3fa475ae66a81895e36b85..0ca82f4b0cd43963afac5f6a8a2449c2342477ad 100644 --- a/19.9m-g014-control/README.md +++ b/19.9m-g014-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-ai-r0.05/README.md b/19.9m-g015-ai-r0.05/README.md index 94e70c8997d60b26de3701346b12e3393021c0f6..e5ba6d4a93918f9a35152ca50e4da37a32e9606e 100644 --- a/19.9m-g015-ai-r0.05/README.md +++ b/19.9m-g015-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-ai-r0.1/README.md b/19.9m-g015-ai-r0.1/README.md index 257d69c5b6fc7be60cb02cae6290d4bd816d1915..96d3d9389e26e8fbca80514e1c7d770fe35a3087 100644 --- a/19.9m-g015-ai-r0.1/README.md +++ b/19.9m-g015-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-ai-r0.25/README.md b/19.9m-g015-ai-r0.25/README.md index c5a1f75996a2564f8ad6e1fafd4d50d879f28869..5626a813d1aaefcbd596de47bb5bd11f49644acc 100644 --- a/19.9m-g015-ai-r0.25/README.md +++ b/19.9m-g015-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-ai-r0.5/README.md b/19.9m-g015-ai-r0.5/README.md index 6253e521df1a25b0caea4a4bb7cc43555c8e27d1..0797e6d9ee4f450b8179a503d2895e83762ed7b6 100644 --- a/19.9m-g015-ai-r0.5/README.md +++ b/19.9m-g015-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-ai-r1/README.md b/19.9m-g015-ai-r1/README.md index 74167b6c2f759ca4f3aa66027fa0c915320c3012..7e4c107a2363a7efe401f3aa9e0129831b44f149 100644 --- a/19.9m-g015-ai-r1/README.md +++ b/19.9m-g015-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g015-control/README.md b/19.9m-g015-control/README.md index b44f45811c5005cf5ebaea9ca1f0087905947331..8628c356017c60641506fa592c359710e4abb1b0 100644 --- a/19.9m-g015-control/README.md +++ b/19.9m-g015-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-ai-r0.05/README.md b/19.9m-g016-ai-r0.05/README.md index cbf6bc3a885f785a0504164f5a278d079061ee38..b6a3bf6c4d0706927d2917e28bca2f04afda6bfe 100644 --- a/19.9m-g016-ai-r0.05/README.md +++ b/19.9m-g016-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-ai-r0.1/README.md b/19.9m-g016-ai-r0.1/README.md index a0f3a775fb73b5555447da18fe2d4429de1a69fd..95d3eb872d80363ecacf46e561543108f765ebfd 100644 --- a/19.9m-g016-ai-r0.1/README.md +++ b/19.9m-g016-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-ai-r0.25/README.md b/19.9m-g016-ai-r0.25/README.md index 8a4022cf0ebc517fec8c9558f20d56ab65ccccb9..d9b072757bf0193bb3d45fdf20b77b0eca5c6754 100644 --- a/19.9m-g016-ai-r0.25/README.md +++ b/19.9m-g016-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-ai-r0.5/README.md b/19.9m-g016-ai-r0.5/README.md index c9c18a01d57b7323a9e404ed66b52240182eb5e8..06345e7dcc51acc22cdd19930e5251aa1a228776 100644 --- a/19.9m-g016-ai-r0.5/README.md +++ b/19.9m-g016-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-ai-r1/README.md b/19.9m-g016-ai-r1/README.md index 54e9be445688baa06d991f0f3e085c9b6d98dd60..b1705b59751ab6a9e1c9f711556dfaad15f53a3c 100644 --- a/19.9m-g016-ai-r1/README.md +++ b/19.9m-g016-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g016-control/README.md b/19.9m-g016-control/README.md index 5195e3c155651d263b0f4eb062a835a1b191806e..f5fd3c2ec5d10a70921a246337e7cb380ce41fd0 100644 --- a/19.9m-g016-control/README.md +++ b/19.9m-g016-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r0.05/README.md b/19.9m-g017-ai-r0.05/README.md index 8d43a9fca0777ac01733d2cbda3073e3f64ab568..1ebcd033f15d9a4758de6a31188b273daf0617db 100644 --- a/19.9m-g017-ai-r0.05/README.md +++ b/19.9m-g017-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r0.1/README.md b/19.9m-g017-ai-r0.1/README.md index 237b5c4fdfb5d575f1311139fe0a15f5a44b8970..e9f7e4c37b7d1e47e9e5af8083bdc55b1fad0a17 100644 --- a/19.9m-g017-ai-r0.1/README.md +++ b/19.9m-g017-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r0.25/README.md b/19.9m-g017-ai-r0.25/README.md index 613ebb843096fddc96ea34a0bbdf84231020efe9..92cf3129992be5fcbd19af9a9a2c1158801cc399 100644 --- a/19.9m-g017-ai-r0.25/README.md +++ b/19.9m-g017-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r0.5/README.md b/19.9m-g017-ai-r0.5/README.md index e952767c2cc4761d9f9560f98659dae81bae5387..1c91bb0ea5cb2093af312adc88c8df2ce330740a 100644 --- a/19.9m-g017-ai-r0.5/README.md +++ b/19.9m-g017-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r1/README.md b/19.9m-g017-ai-r1/README.md index c7ef3a567ed36482c44d4537f8e5f68847de2029..edf354f4a7ed4f4c8562c9e733fb83854cf4197d 100644 --- a/19.9m-g017-ai-r1/README.md +++ b/19.9m-g017-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r16/README.md b/19.9m-g017-ai-r16/README.md index 36afa8b2d176b2c9c782b23a9bb70114aa5caa6a..2727322d1be8468ced1cc74f41501f3dc0e26c19 100644 --- a/19.9m-g017-ai-r16/README.md +++ b/19.9m-g017-ai-r16/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r2/README.md b/19.9m-g017-ai-r2/README.md index ef39f247046330769f860e0b675fd09a551927bd..3f137f96f809ef4ab673980ffa30a163eee20f42 100644 --- a/19.9m-g017-ai-r2/README.md +++ b/19.9m-g017-ai-r2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r4/README.md b/19.9m-g017-ai-r4/README.md index d149fb971f2ed5defd7dc93a7d12db1f8c74d6dc..dee78f244fa29c128abab6fdec61e54ee8e69988 100644 --- a/19.9m-g017-ai-r4/README.md +++ b/19.9m-g017-ai-r4/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-ai-r8/README.md b/19.9m-g017-ai-r8/README.md index 67674836ae0eaf6d79c3bb792c3135bb3dc982a7..23bc9e48e5bfbbfaf4f5178037c4836a21842630 100644 --- a/19.9m-g017-ai-r8/README.md +++ b/19.9m-g017-ai-r8/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-control/README.md b/19.9m-g017-control/README.md index 9d054c00e270f5adfc7f0804b38f6c7e65bb8e83..fb551fbb9438e75517e76ef61887164d6210225f 100644 --- a/19.9m-g017-control/README.md +++ b/19.9m-g017-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g017-human-r2/README.md b/19.9m-g017-human-r2/README.md index 5a46137a396e4fc23261b5b935189b9586558a20..d4d7e1397f923c3faa1aabe8d2bc3d8c84002772 100644 --- a/19.9m-g017-human-r2/README.md +++ b/19.9m-g017-human-r2/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-ai-r0.05/README.md b/19.9m-g018-ai-r0.05/README.md index 023b92cbc34251b2270176d3475f79d8cb0f941f..7938659a608871e9c5f4f71108c6e8192f452ce5 100644 --- a/19.9m-g018-ai-r0.05/README.md +++ b/19.9m-g018-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-ai-r0.1/README.md b/19.9m-g018-ai-r0.1/README.md index 58ebdbb8f6efe4a8645879c15540dde10319a335..1f04a3f99e40b47d9185a4f742b04ffe74ec4a5a 100644 --- a/19.9m-g018-ai-r0.1/README.md +++ b/19.9m-g018-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-ai-r0.25/README.md b/19.9m-g018-ai-r0.25/README.md index 2b83f19213ef3267ef9e8481f68edc68e30314ed..adc879cf70dd1910566ba3b0ad29c420ca782213 100644 --- a/19.9m-g018-ai-r0.25/README.md +++ b/19.9m-g018-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-ai-r0.5/README.md b/19.9m-g018-ai-r0.5/README.md index d185dec41e2442c4e909da86ffed40598efe4de6..940849f1e317fc4ff3d47b31c50bf7ee78eeed29 100644 --- a/19.9m-g018-ai-r0.5/README.md +++ b/19.9m-g018-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-ai-r1/README.md b/19.9m-g018-ai-r1/README.md index 0323e63ef517ac4b197fb4fd145596dbe99876de..180b83aa1f4751eaf7e01bc49d64664419ead753 100644 --- a/19.9m-g018-ai-r1/README.md +++ b/19.9m-g018-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g018-control/README.md b/19.9m-g018-control/README.md index 0143a965ee8affe3cf039ff1ddddbcda871aa880..d2c6ea3f88fb0dfee9d9eff1868d059a8520a858 100644 --- a/19.9m-g018-control/README.md +++ b/19.9m-g018-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-ai-r0.05/README.md b/19.9m-g019-ai-r0.05/README.md index d9b5365e8b21281ac3bf2bf6f90c98fc8978386a..dc0fa7cb4e7baef3ce71e2d5b883285b50f326ff 100644 --- a/19.9m-g019-ai-r0.05/README.md +++ b/19.9m-g019-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-ai-r0.1/README.md b/19.9m-g019-ai-r0.1/README.md index 1c212aa02690a98ce124add141238199a35243a2..2ebc5a91d65b92742787172e0c58cb9e4044ee6b 100644 --- a/19.9m-g019-ai-r0.1/README.md +++ b/19.9m-g019-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-ai-r0.25/README.md b/19.9m-g019-ai-r0.25/README.md index 0f4a75f857445c3d1faef127167c5b7288bca5b3..eecd89f1c097801362d5185bf301ca2b48007aee 100644 --- a/19.9m-g019-ai-r0.25/README.md +++ b/19.9m-g019-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-ai-r0.5/README.md b/19.9m-g019-ai-r0.5/README.md index 7cdb0cc19d3f0deca33f32d07d65dfda70e0260e..d4e9f96584f2714e027fe8e8174ba280cbf5054a 100644 --- a/19.9m-g019-ai-r0.5/README.md +++ b/19.9m-g019-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-ai-r1/README.md b/19.9m-g019-ai-r1/README.md index 7af6448ef3f1bddb48a3af8327fdc23c74824e65..8a4c9f701443333162be1a2118a112455ae1e000 100644 --- a/19.9m-g019-ai-r1/README.md +++ b/19.9m-g019-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g019-control/README.md b/19.9m-g019-control/README.md index f1d1e846da54499f1ed8d3cd8f0aa4ce0f6fd648..9acc7be27c62440eb60c91db51ae7a8c8ae0a0d9 100644 --- a/19.9m-g019-control/README.md +++ b/19.9m-g019-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-ai-r0.05/README.md b/19.9m-g020-ai-r0.05/README.md index 9f6b2594cfa23644c77163e2084c747e27bb37cc..fce047488476d5b7690cca40ddef853b3f1e2f48 100644 --- a/19.9m-g020-ai-r0.05/README.md +++ b/19.9m-g020-ai-r0.05/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-ai-r0.1/README.md b/19.9m-g020-ai-r0.1/README.md index 9e9889566f22738f49dfe6b3cdbf47c5df2d5fac..b40c948f9abf813e754350ee5fe3f7590307c8b2 100644 --- a/19.9m-g020-ai-r0.1/README.md +++ b/19.9m-g020-ai-r0.1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-ai-r0.25/README.md b/19.9m-g020-ai-r0.25/README.md index 095930042002fa473a17fe136b8ae5626ae11e9f..b509d49ed509bfdaf60b6739f688252a918ed793 100644 --- a/19.9m-g020-ai-r0.25/README.md +++ b/19.9m-g020-ai-r0.25/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-ai-r0.5/README.md b/19.9m-g020-ai-r0.5/README.md index 48054091f262542fcab1a12e9ed8a95c32291223..a507cdea11fe9039c8258588cf12a5c298539f58 100644 --- a/19.9m-g020-ai-r0.5/README.md +++ b/19.9m-g020-ai-r0.5/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-ai-r1/README.md b/19.9m-g020-ai-r1/README.md index f28b58524ffab6ad06762709b26bb6bc7ca5e6d2..4a3e7cb820afbd370fe547ccf306e706fb59f685 100644 --- a/19.9m-g020-ai-r1/README.md +++ b/19.9m-g020-ai-r1/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/19.9m-g020-control/README.md b/19.9m-g020-control/README.md index b16beab1f4ea492bf5886056c91e4680c3ce8e7d..8ab7c4862cd3751e506aa6a76a4e84b09cf0226b 100644 --- a/19.9m-g020-control/README.md +++ b/19.9m-g020-control/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 19.9M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -95,7 +95,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/35.8m-f06-ai-removed/README.md b/35.8m-f06-ai-removed/README.md index 5084f7a2d90563f4534a2353e646611e10ef144d..af8aed035dbe00938222243e7f5a6c7ad71c80ee 100644 --- a/35.8m-f06-ai-removed/README.md +++ b/35.8m-f06-ai-removed/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 35.8M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 35.8M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -71,7 +71,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295}, diff --git a/35.8m-f06-web-mix/README.md b/35.8m-f06-web-mix/README.md index 278abd5eb253e4900b3787bcc26b896391b3ec9f..84fa0ba792b1ec98312d5e761c470f390f94fd4d 100644 --- a/35.8m-f06-web-mix/README.md +++ b/35.8m-f06-web-mix/README.md @@ -18,7 +18,7 @@ tags: > **License: [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).** The WildAI code, models and dataset are released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: you may use and adapt them for non-commercial purposes, with attribution, and must share adaptations under the same license. -A 35.8M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of 844 models that measure how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. +A 35.8M decoder-only language model trained from scratch for [*How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text*](https://arxiv.org/abs/2609.40295), one of the paper's models measuring how wild AI-generated web text in pretraining data changes a model's loss. It is a research artifact for studying scaling laws, not an assistant: it has had no instruction tuning or safety training. ## Training data @@ -71,7 +71,7 @@ The nanochat GPT recipe: rotary embeddings, RMSNorm without weights (also on que ## Citation ```bibtex -@article{russell2026aitoken, +@article{russell2026wildai, title = {How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text}, author = {Russell, Jenna and Glickenhaus, Ben and Thai, Katherine and Wieting, John and Iyyer, Mohit and Spero, Max and Emi, Bradley}, journal = {arXiv preprint arXiv:2609.40295},