Notes on language models, agents, and the ways they break.
I write about the inside of LLM systems like decoding, evals, memory, and multi-agent coordination, usually with a benchmark or a worked example to anchor each post. Long-form, occasionally interactive, mostly things I wish someone had written before I had to figure them out.
Build your World Cup bracket
The tournament's on. Rank the 12 groups, choose which third-placed teams go through, and pick the winner of every match to crown a champion.
Open the predictor
Worth reading first
-
LLMs predict the 2026 World Cup
Four frontier LLMs each made a full 48-team forecast for the 2026 World Cup from the same context pack. They put Spain on top, agreed on the top six and almost every group winner, and split on Group D. This post is the record, and I will score them against the real results as the tournament plays out.
-
Why Are the Uyuni Salt Flats Hexagonal?
Walking across the Salar de Uyuni you see a perfect mosaic of metre-wide hexagons stretching to the horizon. The pattern has almost nothing to do with the salt crust itself, and everything to do with porous-media convection happening a few centimetres underground.
-
What an eval suite is, and how to build one
Your first eval suite doesn't need to be huge. Start with checks that catch real regressions: deterministic asserts, a small golden set, and replays from production logs.
-
Why Streaming LLMs Need Attention Sinks
A walkthrough of attention sinks: what they are, why softmax produces them by accident, why naive sliding-window inference collapses without them, and how a four-token reservation lets streaming inference run to four million tokens with no quality loss.