Verified80% confidenceFactTime unknown
PPO（近端策略优化）是OpenAI在2017年提出的一种强化学习算法，成为RLHF的主流选择
4
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedPPO is the core algorithm used by OpenAI in the RLHF training phase of large language models such as ChatGPT.81% similarUnverifiedRLHF 由 OpenAI 于 2017 年前后系统化，最初用于让模型输出更符合人类价值观74% similarUnverifiedRLHF 最初由 OpenAI 用于语言模型的对齐训练，后被广泛应用于图像生成领域70% similarUnverifiedPyGPT supports OpenAI's reasoning models o1 and o3.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/31969API
curl https://kongchang.com/api/v1/knowledge/claims/31969MCP
get_claim(id=31969)