Unverified50% confidenceFactExact time
Under distribution shift, policies trained by predictive learning may systematically produce infeasible solutions.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTraining stability is crucial in reinforcement learning, because policy updates, sparse rewards, and environment randomness all cause curve fluctuations, and some self-distillation methods may exhibit a sudden drop in reward, entropy collapse (the policy converging prematurely to a single behavior), or abnormal gradients midway through training.80% similarUnverifiedReinforcement learning in the post-training stage can improve answer accuracy on both in-distribution and out-of-distribution tasks, but does not consistently improve the validity of reasoning traces.79% similarUnverifiedThe core difficulty of offline reinforcement learning is distribution shift—the learned policy easily deviates from the behavior policy, thereby encountering unseen state-action pairs in the real environment.79% similarUnverifiedEffective reinforcement learning training requires maintaining a certain error rate, keeping the model in a range where it can answer correctly only with effort, which is consistent with the educational psychology concept of the Zone of Proximal Development76% similarUnverifiedThe VICO paper points out that when model capability improves, fixed training tasks deviate from its learning frontier, causing the learning signal to collapse.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153181API
curl https://kongchang.com/api/v1/knowledge/claims/1153181MCP
get_claim(id=1153181)