跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.26584 · v1

Represented but misread: Post-training changes how language models read out what they represent

被表征但被误读:后训练改变语言模型读取其表征的方式

首次提交 2026-11-20 02:43
版本更新 2026-11-20 02:43
雷达观测 2026-11-20 12:52
arXiv ↗PDF ↗

摘要

Does post-training change what a language model represents, or how this representation is read out into an answer? We compared linear probes with the models' own answers, analyzed with signal detection theory, in six open-weight base models and their post-trained versions. For judgments of whether German texts were written by a human or an AI, probes and word n-gram classifiers separated the classes well, but the answers did not and often ordered the texts inversely. Post-training lowered the sensitivity of the answers on average (Δd′=−0.34Δd' = -0.34 across model families; -0.29 across three prompts). It also changed the use of a stated base rate: Base models shifted their log-odds by less than the stated change in prior odds, most post-trained models by more. In two models with stage-wise checkpoints, the change occurred during supervised fine-tuning, model stitching implicated the early layers, and from mid-depth on internal states at positions after the text, not those of the text, transferred the change between checkpoints. Our corpora cannot be shared, so we add a case reproducible from public data, the truth of negated statements. Here, post-training acted in the opposite direction: Probes decoded truth at or near ceiling in every model, the answers of several base models did not use it, and post-training repaired them; by mid-depth the repair resided at the positions after the statement; below it, the representation of the statement contributed, strongly in one of four models. In a training study, adding 2,000 correctly labeled authorship judgments of news texts to otherwise identical fine-tuning data improved the answers on our newspaper and textbook corpora and inverted them on the chats, where they followed a news-trained probe of the base model. Post-training thus changed how a judgment is read out of what a model represents, and a small training signal can set its direction.

作者备注:64 pages, 7+2 figures, 4+6 tables

AI 论文解读

基于摘要 · 非全文审阅

后训练主要改变语言模型如何从内部表征中读出答案,而非改变表征本身,且这种读出方向可被少量训练信号设定。

正在读取账号与研究记录…

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-20 02:43 · 最新观测版本

Represented but misread: Post-training changes how language models read out what they represent

Does post-training change what a language model represents, or how this representation is read out into an answer? We compared linear probes with the models' own answers, analyzed with signal detection theory, in six open-weight base models and their post-trained versions. For judgments of whether German texts were written by a human or an AI, probes and word n-gram classifiers separated the classes well, but the answers did not and often ordered the texts inversely. Post-training lowered the sensitivity of the answers on average (Δd′=−0.34Δd' = -0.34 across model families; -0.29 across three prompts). It also changed the use of a stated base rate: Base models shifted their log-odds by less than the stated change in prior odds, most post-trained models by more. In two models with stage-wise checkpoints, the change occurred during supervised fine-tuning, model stitching implicated the early layers, and from mid-depth on internal states at positions after the text, not those of the text, transferred the change between checkpoints. Our corpora cannot be shared, so we add a case reproducible from public data, the truth of negated statements. Here, post-training acted in the opposite direction: Probes decoded truth at or near ceiling in every model, the answers of several base models did not use it, and post-training repaired them; by mid-depth the repair resided at the positions after the statement; below it, the representation of the statement contributed, strongly in one of four models. In a training study, adding 2,000 correctly labeled authorship judgments of news texts to otherwise identical fine-tuning data improved the answers on our newspaper and textbook corpora and inverted them on the chats, where they followed a news-trained probe of the base model. Post-training thus changed how a judgment is read out of what a model represents, and a small training signal can set its direction.

查看此版本解读 →查看此版本原文 ↗