Unverified50% confidenceFactExact time
DeepSeek探索了GRPO算法，作为对RLHF原始框架的改进尝试
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedAnthropic提出了RLAIF和Constitutional AI，DeepSeek探索了GRPO算法，都是对RLHF原始框架的改进84% similarVerifiedDeepSeek-R1通过在代码和数学两个可验证域上大规模运行GRPO算法取得成功83% similarUnverifiedDeepSeek在训练阶段大量使用GRPO算法替代传统的RLHF来压缩训练成本81% similarVerifiedGRPO由DeepSeek团队系统化提出并在训练DeepSeek-R1系列模型时得到大规模验证79% similarUnverifiedDeepSeek R1通过GRPO算法仅用ORM即可涌现出长程推理能力78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/563140API
curl https://kongchang.com/api/v1/knowledge/claims/563140MCP
get_claim(id=563140)