https://wenhaochai.com/blogs.html Essays, seminar notes, and slides by Wenhao Chai. en-us https://wenhaochai.com/blogs/intro-to-claude-code-and-codex.html https://wenhaochai.com/blogs/intro-to-claude-code-and-codex.html Sat, 13 Sep 2025 12:00:00 GMT A practical introduction to terminal coding agents: memory, permissions, custom commands, MCP, and how Claude Code and Codex complement each other. https://wenhaochai.com/blogs/ode-perspective-on-neural-networks.html https://wenhaochai.com/blogs/ode-perspective-on-neural-networks.html Sat, 26 Jul 2025 12:00:00 GMT Neural networks as numerical discretizations of ODEs: Neural ODEs, ODE and OT Transformers, recurrent-depth models, and what the continuous-dynamics view buys us. https://wenhaochai.com/blogs/transformer-layers-from-online-optimization.html https://wenhaochai.com/blogs/transformer-layers-from-online-optimization.html Tue, 20 May 2025 12:00:00 GMT From linear attention and fast weight programming to Test-Time Training and the Mesa layer: Transformer inference as online optimization over layers and time. https://wenhaochai.com/blogs/intrinsic-dimension-of-your-data.html https://wenhaochai.com/blogs/intrinsic-dimension-of-your-data.html Wed, 15 Jan 2025 12:00:00 GMT ImageNet stores 150,528 pixels per image, yet fewer than 50 coordinates describe where the images actually live. How to estimate intrinsic dimension, and why it matters. https://wenhaochai.com/blogs/bridging-parallel-decoding-with-diffusion.html https://wenhaochai.com/blogs/bridging-parallel-decoding-with-diffusion.html Wed, 30 Oct 2024 12:00:00 GMT Jacobi Decoding generates multiple tokens per step for autoregressive LLMs, and its iterations mirror a discrete diffusion process.