Unverified50% confidenceFactExact time
研究表明CoT提示在参数量低于100B的模型上几乎无效
1
Sources
50%
Confidence
Long-term
Relevance
7/25/2026
First Seen
Sources
Related Claims
Unverified研究发现CoT推理能力只在参数量超过约100B的超大规模模型中自发涌现81% similarVerifiedCoT的涌现性表现为仅当模型参数量超过约1000亿时才能显著提升性能，小模型上几乎无效72% similarUnverifiedCoT效果在参数量超过约100B的模型上尤为显著，体现为「涌现能力」64% similarUnverifiedOpenAI的图表显示，在较低的测试时计算量下，模型对这道题的成功率接近零，只有在极长的推理链条下才偶尔成功64% similarUnverifiedCVE-Zero 测试的样本量小（n=10），作者标注结果为方向性64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/612434API
curl https://kongchang.com/api/v1/knowledge/claims/612434MCP
get_claim(id=612434)