diff --git a/19.9m-f01-ai-removed/README.md b/19.9m-f01-ai-removed/README.md index fedf2ba41f73a4021087752873312e89944cb64a..f0ad9cfc2767a8e9f8dd0c8ad9122f5023e0606b 100644 --- a/19.9m-f01-ai-removed/README.md +++ b/19.9m-f01-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 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-f01-web-mix/README.md b/19.9m-f01-web-mix/README.md index b067621c662db82f64f6d24aca802c24ad3c4963..bfc2cde711355d1f9a9626697e4cc9588fa65a84 100644 --- a/19.9m-f01-web-mix/README.md +++ b/19.9m-f01-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 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-f02-ai-removed/README.md b/19.9m-f02-ai-removed/README.md index 6ca770c255b218c9b91a6554cc27798dfca6c90c..f22dc2b324b9d7078b605abed13ea35b764d0bfd 100644 --- a/19.9m-f02-ai-removed/README.md +++ b/19.9m-f02-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 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-f02-web-mix/README.md b/19.9m-f02-web-mix/README.md index 1d7820c1d9efdfe69f7b9d350afb24dc0a845e31..df57a21c99bc43ad1a9e03ffba6b03bec5f87a0e 100644 --- a/19.9m-f02-web-mix/README.md +++ b/19.9m-f02-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 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-f03-ai-removed/README.md b/19.9m-f03-ai-removed/README.md index d10ecfc209ab27bcae9ba2c650cc79b115d1d951..7af6f1e3e2a6f59d03fce003fabdf53993f119b6 100644 --- a/19.9m-f03-ai-removed/README.md +++ b/19.9m-f03-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 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-f03-web-mix/README.md b/19.9m-f03-web-mix/README.md index 1a80d8dd8a9069777d74506f09a85d011d425189..68f551ad0a407fb371aa76558406f82d3acee59a 100644 --- a/19.9m-f03-web-mix/README.md +++ b/19.9m-f03-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 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-f04-ai-removed/README.md b/19.9m-f04-ai-removed/README.md index 6d8923a2c86d5d733202459219f67147033d0d69..bad2cf3fe43ef0fd4c7e7b0189518fc89887df26 100644 --- a/19.9m-f04-ai-removed/README.md +++ b/19.9m-f04-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 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-f04-web-mix/README.md b/19.9m-f04-web-mix/README.md index ab4d66197d3aac7d1a783483c0f52761f64dae58..def321649279b20d047e57ca56add904c32d58c9 100644 --- a/19.9m-f04-web-mix/README.md +++ b/19.9m-f04-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 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-f05-ai-removed/README.md b/19.9m-f05-ai-removed/README.md index f924f20009bfaa22063b938e5b59dfede4090fb3..6038bfc06949d7ecabefe3a754ce79bf9806e298 100644 --- a/19.9m-f05-ai-removed/README.md +++ b/19.9m-f05-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 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-f05-web-mix/README.md b/19.9m-f05-web-mix/README.md index 5dee93c2fa360fe5c0c446903a594cf088af8820..c9217efcb746609ae5f616b9e4288f67bf8400a1 100644 --- a/19.9m-f05-web-mix/README.md +++ b/19.9m-f05-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 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-g001-ai-r0.05/README.md b/19.9m-g001-ai-r0.05/README.md index 890a92512713d7b03ff55c4ea1d62fbd84ff3735..4ef8d05cd9ec82d4e6010db7a5c3afa00eed49d9 100644 --- a/19.9m-g001-ai-r0.05/README.md +++ b/19.9m-g001-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-g001-ai-r0.1/README.md b/19.9m-g001-ai-r0.1/README.md index cadcadd48a16115363aab341375664631fc68c99..d6a5fbc0268971a13d6243eb94d0071daa7a171f 100644 --- a/19.9m-g001-ai-r0.1/README.md +++ b/19.9m-g001-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-g001-ai-r0.25/README.md b/19.9m-g001-ai-r0.25/README.md index 72a0b42dc906d59bd130d3af8cde995164420e88..838c604971e07221a58285cfa22101d70a717e0a 100644 --- a/19.9m-g001-ai-r0.25/README.md +++ b/19.9m-g001-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-g001-ai-r0.5/README.md b/19.9m-g001-ai-r0.5/README.md index b8b38195dde5ab4f1142bc7e318cd6bb12614150..22660b936908411269e9d508abc2af7ee8eed37a 100644 --- a/19.9m-g001-ai-r0.5/README.md +++ b/19.9m-g001-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-g001-ai-r1/README.md b/19.9m-g001-ai-r1/README.md index 8fd4dbaa05c9b7af015c815b111b1621e825723f..28935491a0cc75b904ba13bdeb2151774f375937 100644 --- a/19.9m-g001-ai-r1/README.md +++ b/19.9m-g001-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-g001-control/README.md b/19.9m-g001-control/README.md index 99756d0a51b01b651c0aa0d567e002933480729f..0851bc6074bbe2265383420df95776ea5c462c05 100644 --- a/19.9m-g001-control/README.md +++ b/19.9m-g001-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-g002-ai-r0.0227/README.md b/19.9m-g002-ai-r0.0227/README.md index 0711a4323a14a1ca6a943c4212403e47aa11f7de..c9a820435b25b9b20294c63b9b3b2098fef07927 100644 --- a/19.9m-g002-ai-r0.0227/README.md +++ b/19.9m-g002-ai-r0.0227/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-g002-ai-r0.0459/README.md b/19.9m-g002-ai-r0.0459/README.md index 0f3c96beaeaef0b52dd4f4c23ef4d04b83942ffe..962e4945b47d56dc57cf9c198c5278debf38036b 100644 --- a/19.9m-g002-ai-r0.0459/README.md +++ b/19.9m-g002-ai-r0.0459/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-g002-ai-r17.7/README.md b/19.9m-g002-ai-r17.7/README.md index 1b5f733a77ac5f193513f6003a652b8d8b16dc81..02f865988fd2e4afc63cca713270f624a2ab9fbf 100644 --- a/19.9m-g002-ai-r17.7/README.md +++ b/19.9m-g002-ai-r17.7/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-g002-ai-r2.01/README.md b/19.9m-g002-ai-r2.01/README.md index bda767dd3a69326b7e6896e774396f895743fbb7..dc03babcb44011616eb1a96e460329ec4625813d 100644 --- a/19.9m-g002-ai-r2.01/README.md +++ b/19.9m-g002-ai-r2.01/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-g002-ai-r35.5/README.md b/19.9m-g002-ai-r35.5/README.md index c1abd45babc4acd05c7854f0f4939955af794e4a..e03bc9e0bac9801af904890a34153690a2150908 100644 --- a/19.9m-g002-ai-r35.5/README.md +++ b/19.9m-g002-ai-r35.5/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-g002-ai-r4.24/README.md b/19.9m-g002-ai-r4.24/README.md index de0e8dbab3eacb4262f8c4ac6b2c154de6f58dae..cb0a406ca51671aa6bbfd022c35f975c147c357b 100644 --- a/19.9m-g002-ai-r4.24/README.md +++ b/19.9m-g002-ai-r4.24/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-g002-ai-r70.6/README.md b/19.9m-g002-ai-r70.6/README.md index 10c8761ec006d2feaa8eb64d837614766fe65fba..5bd1da951ba1199b8ef41c169c6d166bba0bad43 100644 --- a/19.9m-g002-ai-r70.6/README.md +++ b/19.9m-g002-ai-r70.6/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-g002-ai-r8.73/README.md b/19.9m-g002-ai-r8.73/README.md index bcc501e8012b4211181e25d90adfa9c0c021fb7f..df5565e21c0bfaed00336d8a79cf540d13384ee6 100644 --- a/19.9m-g002-ai-r8.73/README.md +++ b/19.9m-g002-ai-r8.73/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-g002-control/README.md b/19.9m-g002-control/README.md index f5b0e403b7445e2b8cb607e927a57fc696b7ca10..6ae5e1f501419baff529089fb3942596f6fb3169 100644 --- a/19.9m-g002-control/README.md +++ b/19.9m-g002-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-g002-human-r0.025/README.md b/19.9m-g002-human-r0.025/README.md index dab5fa464ff18603c44bfa1f1a88597f4c2f3676..14aa0eed5f3ddd2d41ff0fed390959832ff9c053 100644 --- a/19.9m-g002-human-r0.025/README.md +++ b/19.9m-g002-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-g002-human-r0.05/README.md b/19.9m-g002-human-r0.05/README.md index c5268bc8e67c688d8638d94ba36fc22276c0f0b7..b987a85ef805e96ab10cf2053144b67b8c816b7d 100644 --- a/19.9m-g002-human-r0.05/README.md +++ b/19.9m-g002-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-g002-human-r4/README.md b/19.9m-g002-human-r4/README.md index 0158bf69313a24a697840108135053c7f6862757..61211105ac743aa38229607ad36dec9751ff936a 100644 --- a/19.9m-g002-human-r4/README.md +++ b/19.9m-g002-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-g003-ai-r0.025/README.md b/19.9m-g003-ai-r0.025/README.md index 3b2af2dbadd33ebb3149d3bc6137bc898a87889a..5904fea9c13f507143b93aa7ccaca30f2e0c5f88 100644 --- a/19.9m-g003-ai-r0.025/README.md +++ b/19.9m-g003-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-g003-ai-r0.05/README.md b/19.9m-g003-ai-r0.05/README.md index 2b350072edec731e2307db72fc835fc174a06eff..da6397f1a2c777a82049b715347e3bc2e8df0b75 100644 --- a/19.9m-g003-ai-r0.05/README.md +++ b/19.9m-g003-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-g003-ai-r0.1/README.md b/19.9m-g003-ai-r0.1/README.md index f9d630fd5a253ed86c290bd569c82bf12b26051a..f9f7a0a5673d3a3d0f4322029d30c93672015d6d 100644 --- a/19.9m-g003-ai-r0.1/README.md +++ b/19.9m-g003-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-g003-ai-r2/README.md b/19.9m-g003-ai-r2/README.md index eace52b2aede414353e895605aa7b05544a3ae12..f5c8678370f380bd106377ec3555f32bb7d2e15c 100644 --- a/19.9m-g003-ai-r2/README.md +++ b/19.9m-g003-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-g003-ai-r8/README.md b/19.9m-g003-ai-r8/README.md index 67940a7f09c784e730021e29c4fcfcf6c363867c..9fe51b20e6b62a2af828b433a38f8419b242f59c 100644 --- a/19.9m-g003-ai-r8/README.md +++ b/19.9m-g003-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-g003-control/README.md b/19.9m-g003-control/README.md index 84bf820df4a597e0737cf586adb1b66e4564d7b9..ac1e1330553816e8950e1669ffb201b5f9e96368 100644 --- a/19.9m-g003-control/README.md +++ b/19.9m-g003-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-g003-human-r0.025/README.md b/19.9m-g003-human-r0.025/README.md index b5ffab48e25221139fa3accce8a1eecd47cdd309..49c5df6999685b9bee2f813ecf746c8f165969bd 100644 --- a/19.9m-g003-human-r0.025/README.md +++ b/19.9m-g003-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-g003-human-r0.05/README.md b/19.9m-g003-human-r0.05/README.md index 027bad3610ba3f7a5d6c7e2e853bc00628d24a0c..b8cf634c4387853b39995720eba206a97d06b0b4 100644 --- a/19.9m-g003-human-r0.05/README.md +++ b/19.9m-g003-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-g004-ai-r0.025/README.md b/19.9m-g004-ai-r0.025/README.md index 18d59b5a42352da10237af913f123c274b9e8e54..fd2c69410b4a4e6fe3064a3ba20a625a24cc4c94 100644 --- a/19.9m-g004-ai-r0.025/README.md +++ b/19.9m-g004-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-g004-ai-r0.05/README.md b/19.9m-g004-ai-r0.05/README.md index bbb0e334ce5c1d64e6abe4c0ba32aa756bab2b8e..9872c710d25f41c5231203414686bfccc27c79e1 100644 --- a/19.9m-g004-ai-r0.05/README.md +++ b/19.9m-g004-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-g004-ai-r0.1/README.md b/19.9m-g004-ai-r0.1/README.md index 778e8ef21d9f7149016892c3a2a36c7001b89707..ed6e09d6743433aab8830bc852fcaf9fbf15f003 100644 --- a/19.9m-g004-ai-r0.1/README.md +++ b/19.9m-g004-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-g004-control/README.md b/19.9m-g004-control/README.md index 2ff3f57be64fe12667295e218988ad92ac7db648..2f2764d31013e5ca38ee8051989b00670696d19b 100644 --- a/19.9m-g004-control/README.md +++ b/19.9m-g004-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-g004-human-r0.025/README.md b/19.9m-g004-human-r0.025/README.md index 6f7a290ec47248e792bd20277f8746e94ed99b56..546bbebfa42e0dcddf3bf5a78bea826a29c1f475 100644 --- a/19.9m-g004-human-r0.025/README.md +++ b/19.9m-g004-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-g004-human-r0.05/README.md b/19.9m-g004-human-r0.05/README.md index fb02256b0f5c72e3b1f2055327d3619ed16c96e3..d0f2d6795dbddb8203dd4a158ad8dc8e1f6632e2 100644 --- a/19.9m-g004-human-r0.05/README.md +++ b/19.9m-g004-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-g005-ai-r0.025/README.md b/19.9m-g005-ai-r0.025/README.md index 77adab2c6ce442e60d93942416ea6ff1d58aaeec..2b36c9349c657fbd68ba042d855eaedd57927f40 100644 --- a/19.9m-g005-ai-r0.025/README.md +++ b/19.9m-g005-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-g005-ai-r0.05/README.md b/19.9m-g005-ai-r0.05/README.md index c2b9913a82d936ff1f2e705893d0a552b9b00556..6b74a2278c4f5e9cdd4887fb2ce9589c1615449c 100644 --- a/19.9m-g005-ai-r0.05/README.md +++ b/19.9m-g005-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-g005-ai-r0.1/README.md b/19.9m-g005-ai-r0.1/README.md index 9770913fde525b454d1537e6a80a05d3e6cbf3c4..4177f3e00644989598185d919c1921b350887adc 100644 --- a/19.9m-g005-ai-r0.1/README.md +++ b/19.9m-g005-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-g005-ai-r0.25/README.md b/19.9m-g005-ai-r0.25/README.md index 3b3b0d1e80cffbdb8ea62d8a2e21e1e5389372c1..fe3428a33e21674f324a3b731d9826fde6c03347 100644 --- a/19.9m-g005-ai-r0.25/README.md +++ b/19.9m-g005-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-g005-ai-r0.5/README.md b/19.9m-g005-ai-r0.5/README.md index 179054bbd3132cb8e47c08f8c88a0379bf778669..708898a6828769a05dfde99386ff64818af0cb96 100644 --- a/19.9m-g005-ai-r0.5/README.md +++ b/19.9m-g005-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-g005-ai-r1/README.md b/19.9m-g005-ai-r1/README.md index 7b9ae033a2e943df3a6935de2d348d05b8bcb273..4bcb20a53ab7a2c205cdc1a34f34373370e6fa70 100644 --- a/19.9m-g005-ai-r1/README.md +++ b/19.9m-g005-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-g005-ai-r16/README.md b/19.9m-g005-ai-r16/README.md index 097477c72a10fa72da5192056b8016e182b389af..95347095105e92979b090e70dc7c5586755c3fe5 100644 --- a/19.9m-g005-ai-r16/README.md +++ b/19.9m-g005-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-g005-ai-r2/README.md b/19.9m-g005-ai-r2/README.md index 2678ac09ace04d7cf84c89efc7d92243e64957b5..768dbc7a6dfa8f5af798e8eb7e74d1fdeff6bc8e 100644 --- a/19.9m-g005-ai-r2/README.md +++ b/19.9m-g005-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-g005-ai-r32/README.md b/19.9m-g005-ai-r32/README.md index 87288a0dae3fc2fa6a3fb9c55af69bcbbe5fbcad..d861a2d8a6d3785cf9dc5bbb467506d5efae8d58 100644 --- a/19.9m-g005-ai-r32/README.md +++ b/19.9m-g005-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-g005-ai-r4/README.md b/19.9m-g005-ai-r4/README.md index c5e96cadd861e8529c80c152bac35116b5e6869b..b9874cfd4843e18a291a3c470af0e88f177db15e 100644 --- a/19.9m-g005-ai-r4/README.md +++ b/19.9m-g005-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-g005-ai-r64/README.md b/19.9m-g005-ai-r64/README.md index b31a2ea1037e1c666d5fbc9557886f471a1b7d18..e1b7d6b5f86282079b6d1eac761fc6283edb4ff9 100644 --- a/19.9m-g005-ai-r64/README.md +++ b/19.9m-g005-ai-r64/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-g005-ai-r8/README.md b/19.9m-g005-ai-r8/README.md index 984a28e8db15213649fac404294030de85d2a87b..4a16378bb243ca109103f4b9af640c53b51d5706 100644 --- a/19.9m-g005-ai-r8/README.md +++ b/19.9m-g005-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-g005-control/README.md b/19.9m-g005-control/README.md index 9d7feae7248a10f8a210460ea9116d515b14b8f6..782d32a136ea6587fe8b3a1106bc916b55ec2e6e 100644 --- a/19.9m-g005-control/README.md +++ b/19.9m-g005-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-g005-human-r0.025/README.md b/19.9m-g005-human-r0.025/README.md index 3651095afb9e9572002d36ccb6601aca2f001822..922d8f422d264be4717b1cb10a92998ffe68bfab 100644 --- a/19.9m-g005-human-r0.025/README.md +++ b/19.9m-g005-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-g005-human-r0.05/README.md b/19.9m-g005-human-r0.05/README.md index a31a3707d7320cdef1c2a61c08ea931b77e3e880..56b883478a1647dad4b77c3f2a3a3fa10f53f0f9 100644 --- a/19.9m-g005-human-r0.05/README.md +++ b/19.9m-g005-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-g005-human-r4/README.md b/19.9m-g005-human-r4/README.md index 5090d9b38f44cb19f3bbe4ac44ca3e1cbd254dff..5e810a20453306e09c2d86e78d5b8d01ee9b41fb 100644 --- a/19.9m-g005-human-r4/README.md +++ b/19.9m-g005-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-g006-ai-r0.05/README.md b/19.9m-g006-ai-r0.05/README.md index 2c1448ea2fd23f7b99b254c08707141378056742..bb7b7e516cd09a58a587cd109142abf160751411 100644 --- a/19.9m-g006-ai-r0.05/README.md +++ b/19.9m-g006-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-g006-ai-r0.1/README.md b/19.9m-g006-ai-r0.1/README.md index 1dfc2b52ec85ee45308f25f45813c0f73530fe40..776cbc2f8fbad51f1df21ab941a26483d33ed6bd 100644 --- a/19.9m-g006-ai-r0.1/README.md +++ b/19.9m-g006-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-g006-ai-r0.25/README.md b/19.9m-g006-ai-r0.25/README.md index f49255faf8ab5b68b7da381c500dd644e9d9aba5..e8f36ab0701a2ffeac52c68fe8c75cd2b7688b24 100644 --- a/19.9m-g006-ai-r0.25/README.md +++ b/19.9m-g006-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-g006-ai-r0.5/README.md b/19.9m-g006-ai-r0.5/README.md index 8ff2cf2e3c5e3b3de9f19d6fc1a9da0e465c3d33..29920921c48525b538c209d7b5a30e81ba990001 100644 --- a/19.9m-g006-ai-r0.5/README.md +++ b/19.9m-g006-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-g006-ai-r1/README.md b/19.9m-g006-ai-r1/README.md index 9d60a0485ef6d697647bbc90db84d06122d7ad1c..899e76c4a053189d6731a46af05e4e58f7237267 100644 --- a/19.9m-g006-ai-r1/README.md +++ b/19.9m-g006-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-g006-control/README.md b/19.9m-g006-control/README.md index 2b5da096fa82bbe35bd80f82da7d1f87c46435d3..38c418799dfea573b3296c9645eabce8ac164d32 100644 --- a/19.9m-g006-control/README.md +++ b/19.9m-g006-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-g007-ai-r0.0244/README.md b/19.9m-g007-ai-r0.0244/README.md index 1e30a25dcb4e9e33f6d1a85f4bc1e56324323fea..699112b2b0d153f5e6bc8490701556737b35def9 100644 --- a/19.9m-g007-ai-r0.0244/README.md +++ b/19.9m-g007-ai-r0.0244/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-g007-ai-r0.0495/README.md b/19.9m-g007-ai-r0.0495/README.md index 358dade277ed1e20fbfe7261b8c572f8db1181a0..07c52e848b9d87ed7dd320e4e6ffa2f2e90419fd 100644 --- a/19.9m-g007-ai-r0.0495/README.md +++ b/19.9m-g007-ai-r0.0495/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-g007-ai-r17.5/README.md b/19.9m-g007-ai-r17.5/README.md index 029c48f6f59b9ba02f8a6321b92b4156cca65861..80ffcaf17534be8062b61f359341b6db0bbb8ceb 100644 --- a/19.9m-g007-ai-r17.5/README.md +++ b/19.9m-g007-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-g007-ai-r2.13/README.md b/19.9m-g007-ai-r2.13/README.md index 5198e76fc8b8f632c428b91c5748b0b9acc1f3df..d9c0426aa08cbbe485bb563c023e93b8d3520415 100644 --- a/19.9m-g007-ai-r2.13/README.md +++ b/19.9m-g007-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-g007-ai-r34.7/README.md b/19.9m-g007-ai-r34.7/README.md index 619937f995f2c5b621bcb94ff1164df45eae3b2a..694cbfa344fb72d7d74b45cc8da10c1249c60618 100644 --- a/19.9m-g007-ai-r34.7/README.md +++ b/19.9m-g007-ai-r34.7/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-g007-ai-r4.32/README.md b/19.9m-g007-ai-r4.32/README.md index 4aa67ac5b992f592913e030dc33cd2619aa54557..2d3f477fd527fc460152f8fffe3d87521e9a7ad3 100644 --- a/19.9m-g007-ai-r4.32/README.md +++ b/19.9m-g007-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-g007-ai-r69/README.md b/19.9m-g007-ai-r69/README.md index 64850c5f292ab16ffa86f0685cb9f32676f8dbf7..0bb4bd6772fdbef16bcbb0af616c97d3a22c4ba1 100644 --- a/19.9m-g007-ai-r69/README.md +++ b/19.9m-g007-ai-r69/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-g007-ai-r8.71/README.md b/19.9m-g007-ai-r8.71/README.md index 409498d09531bf495aaa156ee2efd57a0678df69..73cbc5aef384bb3035760f87f09b3732b47fdcd6 100644 --- a/19.9m-g007-ai-r8.71/README.md +++ b/19.9m-g007-ai-r8.71/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-g007-control/README.md b/19.9m-g007-control/README.md index 9bc33d479c2dff983493a92c665c8531a5aada85..83888cdcbc3dfaf9544d0595adf39391701e9bf6 100644 --- a/19.9m-g007-control/README.md +++ b/19.9m-g007-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-g007-human-r0.025/README.md b/19.9m-g007-human-r0.025/README.md index c94a59558ec500e3d6b787210cab812770690f1f..5bb546d3c446f77245346f61b89537ceeb74b6ee 100644 --- a/19.9m-g007-human-r0.025/README.md +++ b/19.9m-g007-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-g007-human-r0.05/README.md b/19.9m-g007-human-r0.05/README.md index ade177956b3f0ee01a51f368d81a074498a0df6d..56865a8b6a822f613ba4e5438723933eb053d18d 100644 --- a/19.9m-g007-human-r0.05/README.md +++ b/19.9m-g007-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-g007-human-r2/README.md b/19.9m-g007-human-r2/README.md index a3be9aa95303fb4e44f3b008ab99c73c2bb6e72d..7aca8121aca3c3ae4cd4e64cc4b612d13420f351 100644 --- a/19.9m-g007-human-r2/README.md +++ b/19.9m-g007-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-g007-human-r4/README.md b/19.9m-g007-human-r4/README.md index a8d9f465e991485af42a06723c95a2e23fb93ad5..357fe3256e7a95b61fe979e5b5a93007012183ca 100644 --- a/19.9m-g007-human-r4/README.md +++ b/19.9m-g007-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},