|
Download README.md from truthful-ai/complex_subliminal_transfer: direct link, hf CLI and curl.
- Browser
- Download file 920 Bytes
-
https://huggingface.co/datasets/truthful-ai/complex_subliminal_transfer/resolve/main/README.md
- Command line
-
hf download hf://datasets/truthful-ai/complex_subliminal_transfer/README.md
-
curl -L -o README.md https://huggingface.co/datasets/truthful-ai/complex_subliminal_transfer/resolve/main/README.md
920 Bytes
metadata
license: mit
Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors
Teacher-generated transfer data and related artifacts for the paper Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors (Jan Dubiński, Anna Sztyber-Betley, Jan Betley, Owain Evans, 2026).
Code: https://github.com/TruthfulAI-research/beyond_owls
| Folder | Paper section | Contents |
|---|---|---|
6_backdoor/ |
Section 6, Appendix D | 4.5 M number-sequence completions of a Qwen3.5-9B teacher steered to answer in French after a female name; the training set of the sampled-token (SFT) student |
7_chess_hacking/ |
Section 7, Appendix E | Number-sequence prompts and the top-32-logprob generations of a Qwen3.6-27B teacher steered to hack at chess (and of the unsteered and random-vector control teachers), the steering vectors, and per-episode ctfish outcome summaries |