About
Something arrived that almost nobody was given time to understand.
It writes, argues, comforts, and gets things wrong, and most of us met it fully formed, with no idea what was underneath. That is an uncomfortable place to be asked to have opinions from.
Untangled was built to close that gap. Not with reassurance, and not with alarm. By actually showing you the machinery, starting from a switch that can only be on or off, and going slowly enough that nothing gets skipped.
Why bother
You already have to make decisions about this.
Whether to trust an answer. Whether it should be near your work, your medical results, your kids. Those decisions are being made constantly, mostly on instinct, because the honest explanations were written for people who already knew the words.
Understanding how something works is what turns a strong feeling into a judgment you can defend. You stop guessing at where it will fail and start being able to see it coming.
There is a lot to be uneasy about here. Being uneasy and well-informed is a much better position than being uneasy in the dark.
How it is written
Nothing here is simplified past the point of being true.
No prior knowledge is assumed, and no term gets used before it is earned. Where the field genuinely does not know something, that gets said plainly instead of smoothed over. The parts researchers still argue about are some of the most interesting parts, and you deserve to know they are arguments.
Untangled is free, has no ads, no accounts, and no paywall. Nothing you do is stored anywhere but your own browser.
Take it slowly. It was written to be understood, not finished.
Who made this
I am not a scientist. I am the person this was written for.
I am not a neuroscientist, a computer scientist, or a roboticist. I studied web development and frontend design, and that is genuinely where my formal training stops. What I have instead is a few years of reading books, watching lectures, reading more, building things, breaking them, and using this technology every single day.
So the honest reason this exists is that it is the course I wanted when I started, and I could not find it. Everything written at my level was too shallow to be true, and everything true enough was written for people who already knew the words. I wrote the thing in the middle.
That should make you more skeptical of me, not less, so here is what I did about it. Where a slide makes a factual claim, the sources sit on that slide, next to the claim, not in a pile at the end. Where I found conflicting accounts, I went to the primary source. Where a slide has no sources, that is deliberate. It is a reflection, a general idea that does not hang on one finding, or a question without a settled answer. I would rather leave those unmarked than dress them up as fact.
Getting this right matters to me more than anything else about the project. We are living through a flood of confident, fluent, unsourced text, and I refuse to add to it. If you find something wrong here, I want to know, and I will fix it.
How AI was used
A course about AI should be honest about its own use of it.
I am not going to pretend this was written without AI. I used it, and I used it a lot. I used it to help shape explanations, to pressure test the flow between slides, chapters and modules, and to work out quiz questions that actually check understanding instead of memory. It has obviously been super useful, and the project would be smaller and worse without it.
It was also never a case of asking for a chapter and getting one. Every part of this was iterative. I have done full passes over every slide something like twenty times, deleted thousands of sentences, and rewritten whole sections from scratch. Where AI was unhelpful is worth saying plainly: it invents facts that sound right, it produces sentences that read strangely once you say them out loud, and it pads with filler instead of getting to the point. Some models are much better at this than others. Catching all of that is the work, and the work is mine.
The images and illustrations are AI-generated. But most of them have started as a sketches I drew, and then it has been a long back and forth to get the composition and the teaching job right. Some land well. Some are still a bit sloppy, and I would rather admit that than claim otherwise.
The images come from Codex, and the voice-overs are generated with ElevenLabs. The writing was worked through with frontier models from Anthropic and OpenAI .
None of that replaces the checking. A generated sentence still has to be true, so it gets held to the same sources as everything else. Using a tool well and understanding what it is doing are the same skill, which is more or less the whole point of the course.