Unverified50% confidenceFactExact time
If pass@16 is much higher than pass@1, it indicates that the model's probability distribution contains the correct answer but does not place it at the highest probability position, which is precisely the space where RL can exert its effect through reward signals.
1
Sources
50%
Confidence
Long-term
Relevance
9/7/2026
First Seen
Sources
Related Entities
Related Claims
Unverified讨论者的共识是如果基座模型在pass@16上的成功率为0，RL几乎不可能凭空创造出新能力，但只要存在非零成功概率RL就能将其增强放大79% similarUnverified研究表明，即使能力相对较弱的模型，其pass@10或pass@20的表现往往能够接近强模型的pass@1水平65% similarUnverifiedText Arena使用Bradley-Terry模型作为底层概率框架，模型A获胜的预期概率为1/(1+10^((Rb-Ra)/400))62% similarUnverifiedWhen large numbers of users call the same underlying model with similar prompts, convergence of outputs is almost a mathematical inevitability due to the consistent underlying weights and probability distributions.62% similarUnverified小模型对数据分布更敏感，同样的数据补充量在4B模型上的边际效果往往比大模型更明显61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/873221API
curl https://kongchang.com/api/v1/knowledge/claims/873221MCP
get_claim(id=873221)