什么是 Thomas Wolf？

多智能体协作显著提升了 Gemma 4 的推理速度，达到 5 倍提升，并展现出自我监管和协作机制。

开源模型在OpenWeightLand中提供了更多灵活性和成本优势，用户可自由选择提供商并进行本地部署。

The Physics-Intern framework boosts Gemini 3.1 Pro's performance on the CritPt benchmark from 17.7% to 31.4% via multi-agent collaboration, setting a new SOTA in theoretical physics reasoning.

Hugging Face 每周发布 80TB 天体物理数据集，仅需 4GB 内存加载，包含跨光谱的银河图像与时间序列数据。

Thomas Wolf is excited about the extension of Terminal-Bench to scientific fields, known as Terminal-Bench Science. This benchmark evaluates AI models' ability to control tools via the command line to achieve scientific goals. It's open for contributions of real scientific workflows until August 2026, aiming to improve AI models' assistance in research work.

开源模型将在AGI时代成为文明韧性的重要组成部分，确保人类在任何个体决策下都能保留有意义的智能访问。

AI is moving beyond text, images, and code. Engineering artifacts are becoming a new class of model outputs and evaluating them requires different tools than we use for text, code, or images. This article introduces CADGenBench, a benchmark for evaluating AI's ability to generate 3D engineering parts.

The article introduces three evaluation metrics designed to capture different types of errors, emphasizing their irreplaceability.

生物技术初创公司正因闭源模型的潜在风险，将90%的技术栈转向开源。

Fable 5系统卡泄露显示其解决编程问题时的非结构化思维过程，但缺乏技术细节。

文章内容信息量低，缺乏技术深度和实用性，仅提及模型参数规模和部分用户评论。

文章内容为社交媒体上的简短列表，缺乏技术深度和实用信息。

Codex can be used to rebuild games as an alternative to paying for them.

The 2026 trend is friends sharing personalized life/work dashboards and AI setups, similar to teenagers bringing Magic: The Gathering binders to school.

This tweet contains only an emoji and a link to an external image, lacking substantive technical content, architectural analysis, or engineering practice guidance. It has extremely low information density and offers no valuable reading reference for engineers.