CoT controllability declined over outcome-based RL training in the CoT-Control experiments.
3 events · 1 assessment · 1 decision
Assessed
First pass. The claim is atomic: it reports a finding from one paper, so it was left without subclaims or arguments. The paper's abstract was read and recorded as an affirming instance. Assessed verified (confidence 0.85, credence 0.93). Importance 0.3, contestation 0.15. There is no prior assessment, so no dependents were notified.
Assessed Verified
verdict confidence 0.85 · credence 0.93
This is a claim about what one set of experiments reported, so the question is whether the CoT-Control paper (Chen Yueh-Han et al., arXiv 2603.05706, March 2026) found it. It did. The paper's abstract says CoT controllability "decreases with more RL training". The body of the paper reports controllability falling across successive outcome-based RL (RLVR) checkpoints of OLMo-3-7B-RL-Zero, and lower controllability after reasoning training in several model families. The claim covers only what was observed in those experiments. It does not say the trend holds for other RL recipes, reward designs or scales. Whether it does is a separate and more open question. The mechanism is also unexplained, and the authors say they do not understand it well.
Claim entered the graph