Unverified50% confidenceFactExact time
In that study, changing the left-parenthesis token when writing "kWh" altered the model's final output answer, indicating that fate-turning points are often not the semantically salient parts.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified反事实解释回答的问题是：最少需要改变输入的哪些特征，才能让模型的预测结果翻转66% similarUnverified通过激活修补（Activation Patching）等因果分析方法可以发现，真正影响模型最终输出的内部路径与书面推理并不吻合64% similarUnverifiedWhen models are exposed to the complete timeline, they exhibit stable, hindsight-sensitive answer shifts, tending to provide answers consistent with known outcomes but not reasonable at the decision point.63% similarUnverifiedIn ReAct loops and chain-of-thought prompting, uncertainty markers generated in intermediate reasoning steps are not automatically carried over to the final answer generation step, causing conflict signals to be lost.63% similarUnverifiedThe autoregressive generation mechanism of traditional LLMs determines that they predict token by token in a forward direction, lacking a natural ability to look back.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003507API
curl https://kongchang.com/api/v1/knowledge/claims/1003507MCP
get_claim(id=1003507)