Tag
论文提出 CFRL，一个针对连续流语言模型（如 ELF）的强化学习后训练框架，通过引入 critic 估计中间 token 嵌入的价值，实现可反传的路径梯度，绕过不可微的解码步骤，在四个推理任务上稳定提升模型性能并优于同类基线方法。
AgentLoop is a Product Hunt product that initiates a fresh Codex worker and critic every cycle, enabling automated AI agent loops.
Discusses a new paper from Tsinghua/Zhipu AI on asynchronous reinforcement learning for agents, and notes that their previous GLM-5.2 model uses a critic instead of GRPO.
GLM5.2 reintroduces a critic component for fine-grained variance reduction, suggesting that group-based methods are ineffective for long horizons. The author believes OpenAI and Anthropic already use value models.