Writing
Complicated AI, explained simply.
Technical explanations often preserve the complexity. I write to remove it without removing the truth.
10 August 2026
Part 6 of 6 · LLMs, explained simply
How much AI does your question need?
Why 1 + 1 does not need a research team, and when an AI needs help from a tool.
More writing
Continue reading
When your AI chat needs a fresh page
How old messages can get in the way, and how to know when to start again.
Why does an LLM write one token at a time?
An LLM usually does not reveal a finished answer in one go. It produces a small piece, adds it to the text, then continues.
How does feedback shape an LLM?
The model practises, people or tests compare its answers, and the training process nudges it towards better behaviour.
How does a base model learn to answer us?
A base model can continue text. To make an assistant, people show it what a useful question and answer look like.
How does the internet fit inside an LLM?
Training is like squeezing a huge library into a small suitcase. The pages do not travel with it, but many patterns do.