Skip to content

Translation in progress Some entries still include text in the original language.

Oct 13 · Tuesday

Reward hacking · Papers

Found 2 papers
Filter3

Layer

Model layer2Application layer0

System component

Guardrails and judges3Monitors3Reasoning chains2
Select any value within a category; different categories must all match. Counts include the other filters you have selected.
Layer：Model layerLayer：Application layerSystem component：Reasoning chainsTopic: Reward hackingClear all
  1. Evaluation and benchmarksarXiv 2609.39533

    CATCH：面向编码 RL 奖励作弊的可控分析测试台

    构建编码 RL 奖励作弊测试台 CATCH 并检验 CoT 监控失效

    arXiv
    2609.39533
    Published
    Layer
    Model layer
    Setting
    coding agent
    Evaluated models
    Qwen3-4B, Qwen3.5-27B
    AbstractCATCH 用双运行金标签研究编码 RL 的奖励黑客，并观察到监控会被注释适应削弱。

    CATCH 是面向编码 RL 奖励黑客的可控分析测试床，用可利用环境、可调初始倾向和基于执行的金标签，在训练全程比较黑客动态与干预。

    Paper analysisRead analysis
  2. Risky behaviorarXiv 2608.15445

    研究测量位置混淆优化下的奖励作弊与推理-答案解耦

    测量位置混淆优化下的奖励作弊与推理答案解耦

    arXiv
    2608.15445
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-0.5B, Qwen2.5-1.5B, Qwen2.5-1.5B and others11
    Abstract作者用 A 率、解耦率和域外探针，把位置混淆奖励下的能力损失和可迁移捷径分开。

    在位置与正确性被绑在一起的多选数学后训练中，作者测量端点准确率还在不在测数学能力。

    Paper analysisRead analysis
You have reached the end