课程定位
CS 25 是 Stanford 最受欢迎的 1 学分 seminar 之一,邀请过来自 OpenAI、Anthropic、Google DeepMind、NVIDIA 等机构的讲者。课程没有作业,出勤即可,讲座会直播和录制,适合用来快速扫前沿。
主要内容
- Transformer 架构:注意力、位置编码、变体
- LLM:预训练、对齐、推理、agent
- 生成模型:图像、视频、音频、3D
- 机器人与具身:VLA、世界模型
- 科学应用:医疗、神经科学、材料
- 产业视角:模型、算力、产品
课程大纲 / 课件
每期录像在 recordings 页和 YouTube 播放列表。
先修
基础深度学习和 Transformer 知识(理解 attention),或修过 CS 224N/CS 231N/CS 230。
考核与工作量
1 unit,S/NC,出勤即可;无作业。
适合谁
任何人——想低成本跟踪前沿的学生、工程师、研究者;也适合作为组会/课程补充材料。
学习建议
不用注册也能看 YouTube 录像。建议每期记 3 个 takeaway,不然很容易听完就忘。
相关课程
官方课程描述(英文原文)
Since their introduction in 2017, Transformers have revolutionized Deep Learning, powering large language models (LLM) like ChatGPT and DeepSeek, image and video generation (e.g. Dall-E and Sora), and much more. In CS25, one of Stanford's hottest seminar courses, we delve into Transformer architectures and their diverse applications through classroom discussions and instructor and guest lectures. Topics include LLM, creative uses in art and music, healthcare, neuroscience, robotics, and so forth. We host leading researchers, with past speakers like Andrej Karpathy, Geoffrey Hinton, Jim Fan, Ashish Vaswani, and folks from OpenAI, Anthropic, Google DeepMind, NVIDIA, etc. Lectures are livestreamed and recorded, attracting a global audience with over a million total YouTube views. This is a 1-unit S/NC course, where attendance is the only homework! Enroll via Axess or audit through livestreams or in-person, space permitting. Prerequisites: basic Deep Learning and Transformers knowledge (understanding of attention) or completion of CS224N, CS231N, or CS230. Course website: https://web.stanford.edu/class/cs25/