Unverified50% confidenceFactExact time
Snell等人2024年的研究量化了当模型规模固定时，增加推理算力的边际收益在许多任务上优于同等算力用于继续预训练
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedSnell等人在2024年发表研究《Scaling LLM Test-Time Compute Optimally》，量化了将算力从预训练转移到推理时搜索与验证的权衡关系85% similarVerified推理时扩展（inference-time scaling）通过在推理阶段增加计算量提升模型性能，存在类似预训练的幂律关系76% similarUnverified研究显示推理时计算的扩展律与训练时计算具有相似的效益曲线72% similarUnverifiedThe Scaling Law paper by Kaplan et al. in 2020 revealed that, under a fixed compute budget, there is an optimal ratio between model parameter count and the number of training tokens (the Chinchilla optimal).72% similarUnverified2021年的Chain-of-Thought Prompting论文证明，通过在提示词中加入推理步骤示例，模型的数学推理准确率可提升数十个百分点70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/547003API
curl https://kongchang.com/api/v1/knowledge/claims/547003MCP
get_claim(id=547003)