Unverified50% confidenceFactExact time
The original version used standard supervised fine-tuning (SFT), with data drawn from 139,845 real and synthetic messages across 1,396 conversations.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSFT（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69% similarUnverifiedSFT（监督微调）是ChatGPT等对话模型训练流程的第一阶段68% similarUnverifiedGPT-4、Claude等通用大模型在预训练阶段学习的是互联网上的海量公开文本，倾向于给出通用的回答61% similarUnverifiedVicuna achieved chat capabilities through supervised fine-tuning on ChatGPT conversation data collected from ShareGPT, and its GPT-4 evaluation score once reached 92% of Claude's and ChatGPT's.61% similarUnverifiedSFT源自InstructGPT论文中提出的RLHF流程的第一阶段61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968165API
curl https://kongchang.com/api/v1/knowledge/claims/968165MCP
get_claim(id=968165)