Unverified70% confidenceFactExact time
A 2023 study by Huang et al. pointed out that self-improvement of LLMs without external feedback is not always effective and can easily lead to degraded performance, especially in reasoning tasks.
2
Sources
70%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLonger inference time does not always yield better results; models may fall into the trap of overthinking, and what truly determines output quality is the model's foundational capabilities and degree of alignment.69% similarUnverifiedRecursive Self-Improvement can theoretically cause model capabilities to grow exponentially, but also raises deep concerns in the field of alignment69% similarUnverifiedThe study observed a clear lack of correlation between the correctness of answers and the validity of reasoning traces, with models frequently arriving at correct answers even when the reasoning traces are completely invalid.68% similarUnverified自回归生成机制擅长流畅输出，但容易在需要回溯修正的逻辑场景中出错67% similarVerifiedThere is no reliable positive correlation between an LLM's confidence and the accuracy of its output; it fundamentally lacks metacognitive capability.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/893111API
curl https://kongchang.com/api/v1/knowledge/claims/893111MCP
get_claim(id=893111)