跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2604.09870 · v3

Reading Human Preference from Looped Transformer Internal States

首次提交 2026-04-11 04:00
版本更新 2026-11-18 21:32
雷达观测 2026-11-19 15:04
arXiv ↗PDF ↗

摘要

Show a frozen language model two replies to the same request, one of which a person preferred. Can a small probe on the model's hidden states tell which? On Ouro-2.6B-Thinking, a model that runs its layers four times per token, and Anthropic's HH-RLHF data, the answer is yes, but only faintly. A linear probe picks the preferred reply 61.5% of the time, a small nonlinear network that scores each reply separately 64.9%, and the evaluator trained for version 1 reaches 64.7% on the test split: clearly better than simple rules such as picking the longer reply, but weak. Whether preference is stored as a comparison between the two replies or as a property of each is still open. A linear probe cannot tell, and a network that reads both replies together has an estimated edge of only 0.7 points (95% CI 0.0 to 1.4) over one that scores each reply separately. These numbers hold under three controls: scoring every pair in both orders, splitting the data by pair, and, for the probes, keeping the end of each conversation, where the two replies differ. Version 1's 95.2% and 84.5% were obtained without the first two and do not hold; version 2's corrected relational-probe accuracy, obtained without the third, was four to five points too low. This version replaces both.

作者备注:Version 3 replaces the original manuscript and v2 erratum with a rewritten analysis. It corrects the probe truncation that could remove the replies being compared, tests the original evaluator in both reply orders, and compares linear and nonlinear probes that score replies separately or jointly on pair-disjoint splits

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v3,暂无更早版本可对照。

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。

v3 · 2026-11-18 21:32 · 最新观测版本

Reading Human Preference from Looped Transformer Internal States

Show a frozen language model two replies to the same request, one of which a person preferred. Can a small probe on the model's hidden states tell which? On Ouro-2.6B-Thinking, a model that runs its layers four times per token, and Anthropic's HH-RLHF data, the answer is yes, but only faintly. A linear probe picks the preferred reply 61.5% of the time, a small nonlinear network that scores each reply separately 64.9%, and the evaluator trained for version 1 reaches 64.7% on the test split: clearly better than simple rules such as picking the longer reply, but weak. Whether preference is stored as a comparison between the two replies or as a property of each is still open. A linear probe cannot tell, and a network that reads both replies together has an estimated edge of only 0.7 points (95% CI 0.0 to 1.4) over one that scores each reply separately. These numbers hold under three controls: scoring every pair in both orders, splitting the data by pair, and, for the probes, keeping the end of each conversation, where the two replies differ. Version 1's 95.2% and 84.5% were obtained without the first two and do not hold; version 2's corrected relational-probe accuracy, obtained without the third, was four to five points too low. This version replaces both.

查看此版本解读 →查看此版本原文 ↗