Metraipolis › archive › dc_898b2bb1b7bd
Imprint Reader: From Weight-Update Readout to Behavioral Intervention — arXiv:2609.35261
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35261
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T20:58:38.697Z
- SHA-256 of the archived text
e0416643773546d728996706b98118a8769a9ae5f8c7f07840c3faf0051ccc5e- Stable URL of this page
https://metraipolis.com/documento/dc_898b2bb1b7bd
How to cite
Imprint Reader: From Weight-Update Readout to Behavioral Intervention — arXiv:2609.35261 — arXiv (en), https://arxiv.org/abs/2609.35261 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_898b2bb1b7bd (Metraipolis archive dc_898b2bb1b7bd, sha256 e0416643773546d728996706b98118a8769a9ae5f8c7f07840c3faf0051ccc5e)
BibTeX
@misc{dc898b2bb1b7bd,
title = {Imprint Reader: From Weight-Update Readout to Behavioral Intervention — arXiv:2609.35261},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35261},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_898b2bb1b7bd; sha256 e0416643773546d728996706b98118a8769a9ae5f8c7f07840c3faf0051ccc5e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35261. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
As language models take a growing role in AI development, a natural aspiration is for them to reflect on their own learning process, as humans do, and use that reflection to improve themselves. At the same time, these models have an advantage that human learners lack, since training leaves parameter-level traces that can, in principle, be inspected directly. However, current models cannot decode these traces into an explicit account of what they have learned. To this end, we introduce the \textit{Imprint Reader}, a model trained with \textit{Semantic Mount-and-Read Tuning} (SaRT) to describe frozen weight updates. SMaRT mounts each update onto the Reader and uses an anchor-free meta-query to elicit a natural-language description, while no-change and random-perturbation controls discourage unsupported claims. On held-out updates, the joint Reader reaches judge-based Pass@100 of $2\%$ for knowledge and $16\%$ for behavior. These results demonstrate the feasibility of natural-language readout while pointing to reliability across updates as the next step. Beyond free-form generation, the Reader provides a differentiable proxy for the gap between a specified target behavior and a candidate weight update. Its coordinate-aligned gradients support intervention through MetaEdit. At a $0.5\%$ pruning rate, Reader-guided selection raises measured harmful-prompt refusal from $57.9\%$ to $64.1\%$ under a safety-maintenance target. Using behavior descriptions without target-task training data, MetaEdit increases the frequency of backtracking and sub-goal expressions in mathematical reasoning traces and raises BFCL Overall from $41.69\%$ to $44.60\%$.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.35261 — 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 2026-10-07T20:29:01.815Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).