Unverified50% confidenceFactExact time
OpenAI研究员在2020年提出预训练损失与计算量之间存在幂律关系（Scaling Laws）
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
GPT-5.6与ChatGPT Work发布：AI从对话助手进化为任务执行者
bilibiliTao吃糕7/10/2026
Related Claims
UnverifiedAnthropic研究员最早在2020年提出预训练损失与计算量之间存在幂律关系（扩展律）82% similarVerifiedThe Scaling Law proposed by OpenAI reveals that model performance improves in a predictable power-law fashion as the number of parameters, training data volume, and compute increase.73% similarUnverifiedThe training-time scaling law was proposed by Kaplan et al. in 2020, holding that model capability is primarily determined by parameter count, training data volume, and training compute.68% similarUnverifiedThe FIM (Fill-In-the-Middle) training paradigm was systematically proposed by OpenAI in a 2022 research paper.68% similarUnverifiedOpenAI在InstructGPT论文中证明，少量高质量人类偏好数据对模型对齐的贡献远超大规模无监督训练66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503429API
curl https://kongchang.com/api/v1/knowledge/claims/503429MCP
get_claim(id=503429)