2026.10.05 I built GPT and got more than shitty text What a month of building language models taught me, from counting characters to the calculations inside a transformer.
2026.09.08 The loop, sampling, and stopping Part three of the series: how a simple loop turns predictions into text, how sampling changes the choices, and what makes generation stop.
2026.08.17 A Model without a head is just a body The title itself is an example of why context matters. This article explores how context is built into token representations and how a model arrives at its next-token prediction.
2026.07.31 Your words are not what the model sees Part one of a tour from prompt to model: how tokenizers turn text into token IDs, why character counting is awkward, and how IDs become vectors.
2026.06.25 tinyloop: What makes clankers clank A deliberately small coding agent, you can understand in an afternoon