Unverified50% confidenceFactExact time
When wording is fixed, model outputs are fairly consistent; inconsistency is exposed only when wording changes while meaning stays the same.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExisting evaluation benchmarks tend to fix scenarios and phrasing, allowing models to pass tests by memorizing standard answers while failing in real interactions where phrasing varies—a phenomenon known as benchmark overfitting.78% 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.75% similarUnverifiedThe append-only subtitle strategy (Stable Hypothesis) means that once the model outputs a word it no longer modifies it, avoiding frequent subtitle flickering and rewrites.75% similarUnverified语言的语法结构越简单、惯用模式越统一，大模型在每一步预测中面临的困惑度越低，生成正确序列的概率越高75% similarUnverifiedA model's commitments and memory writes do not change its underlying probability distribution, and its reflection is essentially just generating text that fits conversational expectations rather than truly updating behavioral strategies, so it may repeat the same mistakes.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098441API
curl https://kongchang.com/api/v1/knowledge/claims/1098441MCP
get_claim(id=1098441)