Unverified50% confidenceFactExact time
SFT源自InstructGPT论文中提出的RLHF流程的第一阶段
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
VerifiedInstructGPT and ChatGPT are representative products of the RLHF pipeline.77% similarUnverifiedInstructGPT在2022年首次将RLHF大规模应用于GPT-3，直接催生了ChatGPT的诞生72% similarUnverifiedSFT（监督微调）用人工标注的提示-回答对直接优化模型输出分布，InstructGPT、Alpaca等早期指令微调工作均以SFT为核心70% similarUnverifiedSFT（监督微调）是ChatGPT等对话模型训练流程的第一阶段68% similarUnverifiedRLHF has become the technical cornerstone of products such as ChatGPT and Claude.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/530575API
curl https://kongchang.com/api/v1/knowledge/claims/530575MCP
get_claim(id=530575)