LLMs and Decay

Our conventional response to all media, namely that it is how they are used that counts, is the numb stance of the technological idiot.

I've had mixed feelings about LLMs since I first started using them, with ChatGPT 3.0 or something. I won't claim to have felt anything new - just scroll down any social media app and you'll see other expressions. I've felt everything from awe to terror to contempt to amusement to defiance, etc.

I think I need to stop using them. That's not a particularly easy decision to make. Or, rather, it's easy in the same way that it's easy to decide to eat right and to exercise. The follow-through is somewhat more difficult. The sound of "LLM go brr" is really satisfying for a guy like me who tends to have much more inspiration than time in which to write things.

They make me dumber, and I can't afford to be dumber. A few weeks back, I needed to do something in a repository and got an error back from Claude Code. I don't remember the error - exhausted usage plan, random 500, whatever - but it errored. And I thought, "well, shit, now I can't [do the thing]." And I sat there and I started to wait, and then I realized... I knew how to [do the thing] myself. It was a trivial series of commands. And I was more than a little horrified, and ashamed of myself, and not for the first time.

I'm already concerned about my attention span. I'm prone to doomscrolling, and sometimes I'll do it for hours. I don't know if it's depression or ADHD or what. But if I'm not doomscrolling, it's likely that I'm tabbing back and forth between instances of Claude Code, doing a bunch of things that create a lot of lines of code but that I don't really learn that much from.

And I need to be learning. I intend to move to France in 2028. I intend to be a damn sight more attractive to employers in France than I am now. And failing that, a damn sight more attractive to employers aux États-Unis. Or some other country. Farting around with little proofs-of-concept hasn't really done a whole lot for me so far, and I don't think it's likely to do much for me in the future.

You might suspect that I've been waiting a while to write this post, and that's true. It's part of the core premise behind this blog, which is "Nug Doug: Doing Better." Reading more books and thinking more about what I'm reading. Practicing writing in French (even though, I have to admit, I have used LLMs to check my work there. I'm not really sure of a good alternative). Relearning mathematics conscientiously.

So what's my new approach? Work doesn't change. My employer actually explicitly demands I "default to AI". So this can't be a total break, which makes any kind of change substantially more difficult to achieve. But personally, I still have control.

Well, I'm trying to spend less time programming outside work, for one. We'll see how that goes. But I'm also renewing my LeetCode subscription and practicing DSA.

LeetCode?! I hear you cry in dismay. Yes, LeetCode-ish coding interviews are stupid. But I've found that I actually really enjoy LeetCode. There are no libraries. There's no frontend. There's no CI or CD. There're no tickets, no planning, no prioritization, no meetings, no code review. There's just me, a problem statement, and the syntax and semantics of the language at hand. It's completely unlike actual engineering, and so is actually really refreshing. If I enjoy Wordle, Clues by Sam, logic puzzles, etc, why wouldn't I like LeetCode?

And some dumbass companies still use it as a hiring thing, so it ends up seeming mildly fiscally responsible as well.