Often I get tired of LLMs and the way they’re changing the world, but sometimes they make things a whole lot easier.
I achieved a bunch of things over the last few days that might have taken weeks a year or two ago.
Chess
I learned chess as a child but never really played it until we had a work chess competition for some reason or other. I briefly got into it, forgot about it, and then got into it again after the Queen’s Gambit show on Netflix. Since then I’ve gradually improved with no real focus up to about 1450 Elo on Chess.com. I’ve got a rough idea of a few openings and I’m not completely horrible all the time.
I’d heard that LLMs were pretty good at chess so I wondered whether I could create a custom chess trainer that’d look at my games, work out where I’m weak, and help me achieve some degree of improvement. Given a few fairly simple prompts, I’d exported my entire Chess.com history and had a bunch of analytics on where I’m weak (apparently I’ve a habit of grabbing pawns that I can’t afford to grab in the middlegame). A few more prompts and I’ve now got a trainer that gives me 8 puzzles a day taken from real games that I’ve played (and messed up).
The LLM did have to install Stockfish in order to evaluate positions, and it’s created a small web app because the visualisation on a Claude Code terminal is just not good enough for me to be able to do anything. Code is here, but these days, who needs code?
Boiling water tap
When we moved into our house a few years ago, we were surprised to find that it had a boiling water tap installed. So far it’s proved itself to be fairly handy (though it hasn’t reached the status of the robot lawnmower — something I’d never purchase myself but now couldn’t live without), except whenever it needs maintenance. After I posted my jobs list a few days ago, I asked Claude to help me get on top of my life. It was pretty insistent that I could stop the tap from leaking myself if only I’d order some o-rings and a new filter while I was at it.
This afternoon, I took a bunch of photos of the quirky plumbing arrangement under my sink, and asked Claude to talk me through the fix. After about an hour of struggle, I’d done it. What’s more (and perhaps because this task took place in the real world), I actually think I’ve learned along the way. One part of my fix hadn’t really worked and there was still a leak, but I was able to turn off the water, disconnect the filter, re-connect it properly, and turn everything back on again all by myself.
Fishy screensavers
My one-year-old loves the screensaver that comes with our Roku TV. It has a bunch of fish swimming about and he often shouts “fish” (or his version of that, which sounds more like “fshhh”) when it comes on. A few weeks ago we had a popup on the TV saying that the screensaver was going away and tonight I decided that was unacceptable. Putting an unreasonable amount of computation on the job (Fable 5.2 was made for this sort of thing, yeah?), I connected to the TV, and had Claude build an animation for us.
A few iterations in and we’ve got the kids’ names atop eel houses, sunken pirate ships, fishing hooks appearing and whisking fish away, and big turtles lazily drifting on by. We kept the fish and made it better and more personal!
Socratic Tutor
A few days ago I came across a Jane Street skill called Socratic Tutor — it teaches you “mastery” in a subject wholly through asking questions, letting you reason your way into understanding. I thought I’d take it for a whirl and I’ve found it really interesting. It feels like an interview, except when I swerve the question or answer a slightly different question, it has no qualms in calling me out and telling me to answer the actual question.
I don’t know yet whether I’m really learning anything new by it, but I do feel like my head has taken a beating after an hour or two with it. I gave it the very broad subject of “LLMs” and it’s had me run the whole gamut from quirks of tokenisation through to RLHF.
Conclusion
I can’t pretend any of these things are earth-shattering, and some of them look more like a hammer in search of a nail. None of them would have been impossible in a pre-LLM age.
The boiling water tap would have taken a bunch of Googling and finding helpful YouTube videos, and might have taken about the same amount of time. The fishy screensaver would have been a day to connect to the TV, and then a few weeks to create a much worse version of the screensaver. The chess tutor…a few weeks for the data analysis, probably. I don’t know that I would have built puzzle drills, though I can see that they’re theoretically possible. And finally, the Socratic Tutor… I actually don’t know that I ever could have built that. Maybe that’s the one — the killer use-case for LLMs is in education.
The thing is, it feels like it simultaneously makes education easier and also less necessary. If intelligence is commoditised and sold at 10 cents per master’s level essay, what incentive is there to beat knowledge into your brain? And at this point, I’m not so sure that I’ve learned anything. Maybe my LLM knowledge and my chess have improved, but really it’s too early to say. I did manage to learn about the tap, but I think I could have learned that without an LLM, and I was operating with my hands in the real world. With the others, the illusion of learning is there, but it’s easy to con yourself into thinking you’re learning.
For all their flaws and brilliance and everything else, LLMs enabled me to do a bunch of pretty fun hacky things that made my life slightly better.