Topics

All the plain-English pages, grouped by theme

Here are the plain-English pages in one place. Start wherever fits you: how to use AI sensibly, how change and habits work, how learning and memory work, and what language models reveal about us. For the technical version with mathematics, datasets, and citations, see the technical track.

Using AI

• AI is the engine, you are the pilot — practical, evidence-based methods for getting something reliable out of AI instead of something generic.
• How I work — how I use AI as a tutor and a critic in my own research.

Change and habits

• How change works — the base map: how stuck patterns work on the inside, and what actually helps.
• The brain has no delete key — why a relapse is not a verdict, and what helps if you want out of an addiction.
• When a child can't go to school — what is happening, and how to help without making it worse.
• When your thoughts go in circles — why thinking more doesn't help, and what does.

How learning and memory work

• Learning — a track that gets carved when an answer wins, not a module in the brain.
• How memory actually works — why it is not a file you save and retrieve.

The counterintuitive side of language models

• What language models reveal about humans — the most striking patterns we share with machines, and what they tell us about us.
• LLMs aren't calculators — everything that looks strange if you think they are just fast spreadsheets.
• Why "knows little, believes a lot" — it shows up in language models too, and that reveals what it really is.
• What is a race? — the whole theory explained with water.
• Cross-substrate phenomena — what humans and language models share, and where they diverge.
• Findings and new explanations — new results in language models, and what they explain about us.

Practical use and interventions

• Behaviour design: find the field that blocks — the classic behaviour-design techniques gathered under one mechanism, each mapped to the field it works on.
• Which prompting trick helps your AI? — and when it backfires. Match the technique to where your model sits.
• Fine-tune or prompt? — retraining a model often breaks it; showing it examples works better and costs almost nothing.