跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.20644 · v1

Instrumenting Failure: Scoring Language Model Agents at Every Decision Node in Imperfect-Information Games

首次提交 2026-11-16 15:04
版本更新 2026-11-16 15:04
雷达观测 2026-11-17 13:07
arXiv ↗PDF ↗

摘要

Agent evaluations report task success and rarely say what an agent did at each decision. We score every decision node of three 12B to 14B instruct-tuned language models (LMs) playing five zero-sum imperfect-information games against four opponents on eight properties from decision and game theory, five of them against a CFR+ equilibrium reference. At each node, the LM also states a belief about the opponent's hidden state in a separate call. Without memory and at temperature 0, stated beliefs sit 0.250 to 0.288 in total variation distance from the equilibrium posterior, and where the action responds to the state its agreement with that belief exceeds chance by -0.04 to 0.19. The two calls frame the opponent differently, so the result is a behavioural inconsistency and makes no claim about internal beliefs. Under one prompt and greedy decoding, the LM's action never varies with the state in 10 of the 38 cells where the reference varies its own. Memory lowers the constant-action share in 9 of 13 testable ladders at temperature 0; at temperature 0.2 the cells at 1 fall from 16 to 5 of the 46 informative cells, while the belief results hold. A gate withholds, with a stated reason, any belief score the run cannot support, and seven of the eight non-zero defect rates are attributable to the instrument, a representation, or the LMs.

作者备注:Accepted at the Interpreting Agent Behavior (IAB) Workshop, NeurIPS 2026. 26 pages, 6 figures. Code: https://github.com/mkingopng/instrumenting-failure

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

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 · 2026-11-16 15:04 · 最新观测版本

Instrumenting Failure: Scoring Language Model Agents at Every Decision Node in Imperfect-Information Games

Agent evaluations report task success and rarely say what an agent did at each decision. We score every decision node of three 12B to 14B instruct-tuned language models (LMs) playing five zero-sum imperfect-information games against four opponents on eight properties from decision and game theory, five of them against a CFR+ equilibrium reference. At each node, the LM also states a belief about the opponent's hidden state in a separate call. Without memory and at temperature 0, stated beliefs sit 0.250 to 0.288 in total variation distance from the equilibrium posterior, and where the action responds to the state its agreement with that belief exceeds chance by -0.04 to 0.19. The two calls frame the opponent differently, so the result is a behavioural inconsistency and makes no claim about internal beliefs. Under one prompt and greedy decoding, the LM's action never varies with the state in 10 of the 38 cells where the reference varies its own. Memory lowers the constant-action share in 9 of 13 testable ladders at temperature 0; at temperature 0.2 the cells at 1 fall from 16 to 5 of the 46 informative cells, while the belief results hold. A gate withholds, with a stated reason, any belief score the run cannot support, and seven of the eight non-zero defect rates are attributable to the instrument, a representation, or the LMs.

查看此版本解读 →查看此版本原文 ↗