- Finding Frequencies Without an FFT: Sinusoids as Autoregressive Models
- A Token-by-Token Guide to Prompt Caching
- Probability of an LLM Trajectory: Starting from the Chain Rule
- Distilling a Bigram
- Build a Bigram Model in 10 Minutes
- Memory to Train a Transformer
- Estimating Layers in Transformers
- Why KV Cache and Not QKV Cache?
- Is RL Secretly Supervised Learning?
Currently Reading — LLM-Generated Notes
Also: CEO Hour — a timed thinking practice you can run on your own.