stanford

Stanford researchers retested 50 widely cited findings about LLMs using 20 new models and found that 44% no longer held, 34% had significantly weakened, and only 16% still held, suggesting that LLM research findings may have a “shelf life” of only about 15 months.

The tweet shares Andrej Karpathy's view from a Stanford class: with tools, memory, state and loops, a model becomes much closer to a programmable computer — and an agent is more than just an LLM with a prompt.

A Chinese-language self-study tutorial for Stanford’s CS329Z “Engineering AI Agents” course has been released, featuring 19 lectures and 4 hands-on exercises covering LLMs, RAG, tool calling, memory, multi-agent systems, and evaluation, with support for offline HTML reading.

Recommend Stanford’s free course CS146S, “The Modern Software Developer” (taught by Mihail Eric), and introduce the accompanying open-source project NamiAgent—a Chinese-language tutorial on building an AI agent from scratch, progressively implementing core capabilities such as tools, permissions, context, Skills, MCP, sub-agents, and memory, complete with web-based visual debugging.

Recommend 10 AI and machine learning courses freely available to the public through Stanford University in 2026 (such as CS221 Artificial Intelligence, CS229 Machine Learning, CS230 Deep Learning, CS234 Reinforcement Learning, CS236 Deep Generative Models, and CS336 Building Large Language Models from Scratch) for self-improvement.

A Stanford CS 312 "Deep Learning Alchemy" course shifts the burden of theory construction onto students (and their Claude assistants), having them empirically probe transformer components via ablations on a 614M-token depth-8 transformer — prompting reflection that maybe memorized ML knowledge was never that critical.

A recommendation of Peter Thiel's 41-minute business lecture at Stanford, covering how to build a monopolistic company, the dangers of commoditized competition, and the key logic behind the success of Google, PayPal, and Facebook.

A recommendation for Stanford's CS329Z 'Engineering AI Agents' course, which systematically covers the core of agent engineering: LLM fundamentals, RAG, tool calling and function calling, framework orchestration (DSPy, LangGraph), evaluation and reliability, and more — signaling that agent development is shifting from an API-patching craft into a serious engineering discipline.

Stanford NLP congratulates alum Melvin Johnson on becoming a VP of Research at Google DeepMind, with a mention of a discussion about Gemini 4 Argon.

AI帮助学生在作业中拿到接近满分，但在线下期末考试中却集体暴露问题。一位斯坦福教授表示，学生用AI完成作业后不再参加答疑，却无法通过线下期末考试，因此院系重新定义助教职责，取消传统答疑，要求学生一对一向助教讲解每份编程作业以证明真正理解代码。

A long-time CS336 fan praises the course's empirical focus and expresses excitement for Stanford's new CS312 'Deep Learning Alchemy' class, announced by Tatsunori Hashimoto, which aims to teach the messy, empirical side of frontier deep learning research.

斯坦福大学新开设 CS 312「Deep Learning Alchemy」课程，由 Tatsunori Hashimoto 和 Suhas Kotha 主讲，聚焦深度学习中以经验为主的知识——超参数缩放、优化器不变性（μP、Muon）、损失景观与训练稳定性等，全部作业代码、测验和讲义公开，可视为 CS 336 的续作。

Lee Robinson 发布了在斯坦福 CS146S 的讲座视频，讲解 Grok Bot 的工作原理，涵盖 Always-on Agent 设计、harness 设计、上下文工程以及如何保持主 agent 轻量等内容，值得一看。

推文推荐斯坦福大学的免费 LLM 课程（含中英双语字幕），称其为讲解 ChatGPT 与 Claude 底层构建原理最清晰、最深入的课程，能帮助学习者真正理解 tokenization、Attention 等核心机制而非停留在记住名词。

Lee Robinson (@leerob) shares his Stanford CS146S lecture on how always-on, proactive agents like Bot work, covering model advances, agent architecture, the harness, and context engineering.

斯坦福大学2026年秋季新课CME 295《Transformers与大语言模型》开放选课，第二讲笔记发布，系统梳理模型家族、缩放定律、MoE、架构优化、解码策略与提示范式等LLM核心主题。

A tweet recommending a complete Stanford AI course video that teaches how to build and automate anything with AI, urging viewers to spend two hours watching it.

Stanford's CS224V "Agentic AI" course, led by Monica Lam's team, builds an anti-hallucination knowledge stack covering RAG, knowledge curation (STORM), formal methods, and meta-agents to turn unreliable LLMs into trustworthy, accountable agents. The first two lectures' slides are now publicly available.

Stanford's Monica Lam team has released CS224V "Agentic AI," a research course focused on anti-hallucination and accountability. Using retrieval, declarative specifications, theorem provers, and rigorous evaluation, it turns unreliable LLMs into trustworthy agents, with projects accounting for 65% of the grade.

Lior announces the MMBU Challenge, a collaboration with Stanford to test whether vision-language models can truly recognize and localize biomedical images rather than guessing answers, featuring 3 tracks, $100K+ in prizes, running Oct 1 to Dec 31 with registration closing today.