Unverified50% confidenceCautionExact time
Tokenization asymmetry is a confounding factor: single-digit answers and two-digit answers occupy different numbers of tokens in the tokenizer, and directly comparing probabilities introduces a systematic bias.
1
Sources
50%
Confidence
Long-term
Relevance
9/25/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEvery character in a prompt affects the probability distribution of all subsequent tokens, which is why subtle wording changes can lead to drastically different outputs.66% similarUnverifiedSystematic magnitude errors are more insidious and more dangerous than random errors, because the answers appear numerically plausible in structure yet deviate by an order of magnitude.65% similarUnverifiedtop-1%贪婪解码准确率只考察单token预测的一致性，无法捕捉量化对长序列生成质量的真实影响，微小偏差会在自回归生成中累积65% similarUnverifiedThe responses of large models such as ChatGPT, Claude, and Perplexity are stochastic, and answers to the same question may differ across different times and contexts.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/949186API
curl https://kongchang.com/api/v1/knowledge/claims/949186MCP
get_claim(id=949186)