Metraipolis › archive › dc_8350eaa342b2
Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect — arXiv:2611.10110
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.10110
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-10T18:40:44.505Z
- SHA-256 of the archived text
2fc1d64ade976062de0cad615bc923c1a0b457f93f973d4ca39e2aab149818f6- Stable URL of this page
https://metraipolis.com/documento/dc_8350eaa342b2
How to cite
Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect — arXiv:2611.10110 — arXiv (en), https://arxiv.org/abs/2611.10110 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8350eaa342b2 (Metraipolis archive dc_8350eaa342b2, sha256 2fc1d64ade976062de0cad615bc923c1a0b457f93f973d4ca39e2aab149818f6)
BibTeX
@misc{dc8350eaa342b2,
title = {Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect — arXiv:2611.10110},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.10110},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8350eaa342b2; sha256 2fc1d64ade976062de0cad615bc923c1a0b457f93f973d4ca39e2aab149818f6},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.10110. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We study homeopathic learning, a phenomenon where poisoned training data with little effect on its own can create a hidden backdoor once clean, contrasting data is added. In our experiments, a model fine-tuned only on a "poison" dataset associated with a trait T (such as speaking German or being misaligned) does not learn T. Surprisingly, adding similarly formatted "clean" data not associated with T causes the model to learn T. This effect requires that the poison data have an identifying "trigger" feature that is absent from the clean data. We demonstrate this effect in a variety of settings: subliminal learning of backdoor behaviors, inoculation prompting of reward hacking transcripts, generalization from archaic bird names, and an MNIST classifier with mislabeled data. We propose a simple regression model as an explanation. In the poison data, the task and trigger always co-occur, and regularization favors attributing T to the task. Clean data without the trigger causes T to be attributed to the trigger instead. Consequently, the same training examples can have little effect as part of one training set but create a misalignment backdoor in another. This finding exposes a potential limitation of safety methods that audit or filter training data in isolation. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.10110 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-10T18:37:37.478Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).