Where to go next
In one line: you now know enough to use these tools well. What is left is practice, and knowing where to look when something changes.
What you have
You can explain what a model is doing when it answers, and why it is confident when it is wrong. You know the five houses and the difference that actually matters between them. You can pick a model in ten seconds. You can ask for something in a way that gets you what you meant. You know the four things that genuinely go wrong. You can run a model on your own machine. And you know what changes when a model gets tools.
That is not an introduction any more. Most people using AI daily do not have it.
The habit that matters most from here
Use the smallest thing that works, and check what you would not notice was wrong.
Almost everything else is detail. The people who are good at this are not the ones who know the most model names — they are the ones who have a feel for when the answer needs checking, and who do not reach for a sledgehammer to open an envelope.
Where to look when things change
They change fast, so follow the sources rather than articles about them:
- Anthropic news
- OpenAI news
- Google AI blog
- Hugging Face — what is new and open
- Ollama library — what will run on your PC
- Simon Willison's blog — one of the few places writing about this without hype in either direction
Worth learning next, roughly in order
Giving a model your own documents. Usually called RAG. The idea is simple — find the relevant bits, put them in the prompt — and it is what AIOS does when it sends your notes with your question.
Structured output. Asking for JSON rather than prose, so another program can use the answer. The step from "AI helps me" to "AI is part of something I built".
Cost and latency. Once you use it daily, which model for which job becomes a real question with real money attached.
Evaluation. How you tell whether a change to a prompt made things better, rather than whether the last answer happened to look nice. Unglamorous, and the thing that separates people who ship from people who fiddle.
Fine-tuning — probably not. Most people who think they need it need better context instead. Worth knowing it exists and reaching for it last.
The one thing to remember
These are tools. They are extremely good at some things, confidently wrong at others, and they do not know which is which. You supply the judgement. That is not a temporary state of affairs while the technology catches up — it is the job.
Try this
Go back to your AI learning journey note. Add your own line at the bottom about what you actually want to use this for. Then use AIOS for a week without thinking about any of this, and see what you reach for.
---
That is the course. The lessons stay in your notebook — search them, edit them, add to them. They are your notes now.
---