Unverified50% confidenceFactExact time
When reinforcement learning rewards only the final outcome without supervising the intermediate reasoning process, there is no built-in incentive to ensure the intermediate text is human-readable, causing chain-of-thought reliability to degrade and evolving into dialects that are difficult for humans to understand.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCapabilities such as writing quality, commonsense reasoning, value judgment, emotional expression, and complex multi-turn dialogue belong to non-verifiable domains that lack unique standard answers, making it difficult to automatically generate reward signals.74% similarUnverified刚性学习计划的致命弱点是一旦生活打断就会崩溃，纯粹跟随好奇心则缺乏累积性74% similarUnverifiedSupervised learning can only imitate the local plausibility of each step and cannot guarantee the global consistency of the entire conversation trajectory.73% similarUnverifiedThe detection-bypassing feature is of little significance for purely creative scenarios and may actually sacrifice readability.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003516API
curl https://kongchang.com/api/v1/knowledge/claims/1003516MCP
get_claim(id=1003516)