Unverified50% confidenceFactExact time
Mechanistic interpretability analysis found that for the same math problem, the translated and localized versions often have markedly different internal reasoning trajectories activated in the model, even when final accuracy is close.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
Unverified与传统可解释性方法（如特征重要性、显著图）不同，机制可解释性追求对模型内部计算的完整逆向80% similarUnverified研究表明，即使使用相同的基础模型，引入迭代反思机制后在代码生成、数学推理等任务上的准确率可提升20%-40%75% similarUnverifiedAfter current mainstream large language models form incorrect premises early in their reasoning, subsequent steps tend to rationalize those premises to maintain consistency rather than backtracking to correct them.74% similarVerifiedReasoning models perform internal Chain-of-Thought before returning final results, consuming large amounts of tokens and computation time in exchange for deeper understanding and lower error rates.73% similarUnverified多项实验表明，模型的表面推理与其实际决策机制之间存在显著偏差，通过激活修补可发现真正影响输出的内部路径与书面推理不吻合73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1078819API
curl https://kongchang.com/api/v1/knowledge/claims/1078819MCP
get_claim(id=1078819)