ClementDelangue

Most people training agentic LLMs with RL right now have a silently broken training loop and have no idea. Here's the trap: single-turn RL works beautifully. Clean curves, sane rewards, everything converges. Then you add tools so the model can act mid-rollout, and things get weird. Loss spikes for no reason. Eventually a shape-mismatch error. The culprit: every time you parse the model's output to detect a tool call, then re-tokenize the updated conversation for the next turn, you're rolling the

Between May 2024 and May 2026, the performance of local open-weight AI models on laptops has improved more than twice as fast as Moore's Law.

文章讨论了用户倾向于默认选择前沿AI模型的现象，而非主动挑选最优模型。

The content is a social media post and does not provide technical depth or practical information.

Andrew Chen分享了他对Reachy Minis的体验，强调其快速构建和活跃社区，但内容缺乏深度和实用信息。

The tweet calls for more attention to Arcee, a U.S.-based open-source AI model company, but lacks depth, novelty, and practicality, serving only as a simple recommendation.

ClementDelangue 请求 Josefchen 将模型和数据集添加到 HF。

ClementDelangue posted a photo on X, asking if Jason had just arrived in Miami.