Metraipolis › archive › dc_ffc6de1ad8ce
Post-Training Leaves Behavioral Shadows on Unrelated Decisions — arXiv:2609.29233
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.29233
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T04:27:21.646Z
- SHA-256 of the archived text
4fe5d29fee0c46c49f39225acbf083cde207b437e6465b8283a4b84f532a5840- Stable URL of this page
https://metraipolis.com/documento/dc_ffc6de1ad8ce
How to cite
Post-Training Leaves Behavioral Shadows on Unrelated Decisions — arXiv:2609.29233 — arXiv (en), https://arxiv.org/abs/2609.29233 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ffc6de1ad8ce (Metraipolis archive dc_ffc6de1ad8ce, sha256 4fe5d29fee0c46c49f39225acbf083cde207b437e6465b8283a4b84f532a5840)
BibTeX
@misc{dcffc6de1ad8ce,
title = {Post-Training Leaves Behavioral Shadows on Unrelated Decisions — arXiv:2609.29233},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.29233},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ffc6de1ad8ce; sha256 4fe5d29fee0c46c49f39225acbf083cde207b437e6465b8283a4b84f532a5840},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.29233. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We find that language models can transfer capabilities through task-unrelated text. Post-training typically improves language models using task-specific data. Prior work on subliminal learning shows that information about these updates can pass through unrelated generations, but has largely focused on traits or preferences using extensive teacher outputs. We introduce Active Taskless Distillation (ATD), which achieves capability transfer using only a single word from the teacher per prompt. ATD probes the behavioral shadow of post-training by selecting prompts where the teacher and student's shared public ancestor is nearly indifferent between two ordinary words. A student initialized from this ancestor learns solely from the resulting prompt-word pairs, without target-task examples, teacher logits, or teacher parameters. In the primary coding experiment with Qwen2.5-1.5B, 5,664nses yield a 5.34 pp gain on HumanEval+ over an exact nuisance-matched control thadisrupts prompt-resperiments showtransfer in scientific knowledge, commonsense reasoning, and reading comprehensins across additional model generations, sizes, and families. Functional analyses show that the learned sid composable, andthat its strength tracks the teacher's update strength. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.29233 — 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 2026-10-08T03:50:58.024Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).