A Brilliant end

I was gifted a Brilliant.org subscription for 2019's Christmas, and for a time it was good. I worked through the courses and learned many new things. Hat puzzles, formal logic, contest maths. Group theory, applied statistics, machine learning. All good stuff.

They added more courses. I learned more from their contributing authors — Sabine Hossenfelder's quantum mechanics course, MinutePhysics's special relativity, and one that doesn't name the author but taught me how photovoltaics actually function. And for a time it was good.

But the UI changed. It became harder to use. Tab no longer goes between text fields. Just like Duolingo, it's gained "cute" animations that take me out of a flow state. The app used to allow courses to be downloaded for offline use — sometimes that didn't work, but usually it did. Now? There's no option to download, and each lesson takes ages to load; so long in fact that, even on home WiFi, I see popup messages asking me if I'd like to give up waiting for the course to load, and that can happen multiple times before I can start.

Sometimes lessons were updated or courses replaced. That's not a bad thing in itself, but what was a problem was that the progress bar for each course was no longer consistent across the UI, showing different levels of progress in different places.

Now, though… well, now I'm stopping. My subscription has run out, and I won't be renewing it, at least not for some time. The new courses I've been seeing recently have been focussing on the low end — I absolutely do not benefit from yet another "here's how triangles work!" course, such content is an absolute waste of my time at this point. Where I'm stuck now is the hard stuff — abelian groups, curls, wedge products, Hessians and Laplacians — which seem as hard now as it ever was. I've reached the point where I don't understand the notation, and without those specific courses being expanded, just re-reading the existing content isn't going to help me understand.

Actually, it's worse than that. At this point, ChatGPT — the free tier — knows more than I do about these topics. This isn't to claim the LLM is in any objective sense good at this level (some of the problems are such that it's possible for me to verify the AI is wrong even though I can't generate any answers in the first place, kinda like P!=NP), I'm just asserting a relative performance between it and me: it knows more than I, personally, know. And this is the worst the AI will ever be.

I may come back to Brilliant again in the future. But I suspect not — by the time they release the content I'm interested in, the LLMs will be teaching it for free… or just directly doing the tasks for which I wanted to learn these topics.

All Good Things…

Tags: Education, Learning, MOOC, STEM

Categories: AI, Opinion, Software, mathematics

© Ben Wheatley — Licence: Attribution-NonCommercial-NoDerivs 4.0 International