Stanford CS25 lecture notes updated: From Tokenization to Transformers
tengyuma · x · 2026-08-19
The instructor for Stanford's CS25 course (Transformers United) announced an update to the lecture notes. These notes are regarded as one of the clearest end-to-end explanations of how models like ChatGPT and Claude are built.
The curriculum covers a wide spectrum:
- Basics: Tokenization and BPE
- Core: Transformer Architecture
- Pipeline: Training Pipeline and Next-token Decoding
The notes are recommended for everyone from beginners to engineers shipping agents daily. The instructor notes that the field moves fast, so the notes remain a work-in-progress, and feedback is welcomed.
More from Research
- New Framework Enables Complex Physics Simulation for 3D Gaussian Splats — Scobleizer · 2026-08-19
- GV's Dave Muni on AI for Science trends and Qualcomm's acquisition of Modular — ajratner · 2026-08-19
- MLLM-Guided Semantic Correction for Text-to-Video Generation — kwangmoo_yi · 2026-08-19
- The Next AlphaFold Won't Design Drugs — It Will Find the Right Targets — julia_kiseleva · 2026-08-19
- Paper: Silly Rules Help AI Understand Human Normative Systems — ghadfield · 2026-08-19
- Train any neural net inside ComfyUI: workflows for autoencoders, UNets included — enspiralart · 2026-08-19