Expired50% confidenceFactExact time
The Process Reward Model provides fine-grained feedback on the Agent's intermediate reasoning steps, alleviating training instability caused by sparse rewards.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/10/2026
First Seen
Valid until: 10/8/2026(expired)
Sources
Related Claims
Unverified训练SWE-bench智能体的先进方法包括拒绝采样（Rejection Sampling）、DPO（Direct Preference Optimization）以及过程奖励模型（Process Reward Model）72% similarUnverified针对工具调用的强化学习训练通常采用执行反馈机制，将实际执行的成功/失败信号作为奖励，可大规模自动化71% similarUnverifiedSkill本质上是通过模拟少样本学习效果，在不修改模型权重的前提下，通过提示模板约束模型的输出分布70% similarUnverified早期强化学习领域的Agent依赖人工设计的奖励函数在封闭环境中学习策略70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/413960API
curl https://kongchang.com/api/v1/knowledge/claims/413960MCP
get_claim(id=413960)