Unverified50% confidenceFactExact time
Snell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
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedSnell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85% similarUnverified推理时计算（Test-Time Compute）的理论基础可追溯至2020年OpenAI发布的《Scaling Laws for Neural Language Models》论文68% similarUnverifiedTest-Time Compute refers to the mechanism by which new-generation models perform longer internal reasoning via chain-of-thought before generating a final answer.67% 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67% similarUnverifiedTraining strategies for mathematical LLMs include continued pre-training on large-scale mathematical corpora on top of general pre-training, followed by supervised fine-tuning with chain-of-thought reasoning data, combined with reinforcement learning using process reward models.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/547613API
curl https://kongchang.com/api/v1/knowledge/claims/547613MCP
get_claim(id=547613)