Unverified50% confidenceFactExact time
Lanham等研究发现较大模型的结论更不容易被CoT中的修改所影响，暗示更强大的模型反而更多依赖内部隐式计算而非显式的文本推理
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
UnverifiedTurpin et al. (2023)的研究证明当提示中植入误导性偏见时，模型会在CoT中给出与真实计算无关的解释，即不忠实推理链问题72% similarUnverifiedExperiments show that Coconut can achieve results comparable to or even better than traditional CoT on certain logical reasoning and planning tasks with fewer computational steps.69% similarVerifiedResearch indicates that hallucination rates do not have a linear negative correlation with model scale, and larger models sometimes produce harder-to-detect misinformation in a more fluent manner.68% similarUnverifiedSmaller models are more prone to embedding collapse than larger models, due to reasons including limited parameter capacity, a tendency toward optimization shortcuts, and amplified anisotropy issues.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/782959API
curl https://kongchang.com/api/v1/knowledge/claims/782959MCP
get_claim(id=782959)