Magistral: Reasoning without Traces (Feb 2026)
Notes on Magistral, Mistral's reasoning model, and what it means to reason without distilled traces.
A Lot of Mistakes: What I Learned in 2025 (Dec 2025)
After getting a lot of things wrong — and some things right
The Thought - Action - Observation Cycle (May 2025)
How agents think, act, and observe — the loop at the heart of agentic LLMs.
Tools in LLMs - What, How, Why? (May 2025)
What tool use in LLMs actually is, how it works, and why it matters.
LLMs Under the Hood (May 2025)
Messages, special tokens, and chat templates — how a conversation actually reaches the model.