Hynek SchlawackApr 12

RE: https://mastodon.social/@bcantrill/116393325512331807

“As for the artifact that [Garry] Tan was building with such frenetic energy, I was broadly ignoring it. Polish software engineer Gregorein, however, took it apart, and the results are at once predictable, hilarious and instructive: A single load of Tan’s "newsletter-blog-thingy" included multiple test harnesses (!), the Hello World Rails app (?!), a stowaway text editor, and then eight different variants of the same logo — one of which with zero bytes.” 😂😭

Show threadHynek Schlawack

I love that article because while I find LLMs more useful than the average fediverser, my most observed and least talked about pet problem is that they allow everyone get their first idea implemented.

You don’t need any skills whatsoever to get started so you can’t judge whether the idea is worthwhile at all and you don’t get any feedback on the feasibility of the design until you run out of tokens. Anything can be brute-forced. The FOSS slop wave is just a sub problem of this.

Apr 12 at 7:44pmMona for iPhone
Show threadJudy2kApr 12
@hynek Fortunately, my first ideas are always excellent. Sadly, it can take a while for everyone else to catch up so that they can appreciate them.
Show threadHynek SchlawackApr 12
@judy2k True-ish
Show threadJudy2kApr 12
@hynek Which is better than True!
Show threadBen NuttallApr 12
@judy2k @hynek I haven't had a bad idea yet
Show threadDavid R. MacIverApr 12

@hynek I've also found that I need to check to see whether it actually implemented my idea, or if it implemented some half-arsed version of it that misses the key things I explicitly included that were required for it to work properly.

(I also find LLMs extremely useful, but if I want the thing to actually work I usually use them with me in the loop pretty heavily and I still miss stuff)

Show threadDavid R. MacIverApr 12
@hynek TBH though I'm really surprised at how bad the site in the story is. Even when Claude is most on its bullshit I've never seen results close to this bad.
Show threadalphaApr 12
@DRMacIver @hynek I think this is in no small part due to the user of the tool.
Show threadHynek SchlawackApr 13
@DRMacIver I think once you allow entropy to accumulate it only gets much worse. I’ve found LLMs most useful when they're forced to operate in a tight harness of good design with constant oversight. Mr Tan is trying to prove a point so he's doing the opposite on purpose and entropy tends to go exponential.
Show threadTin TvrtkovićApr 13
@DRMacIver @hynek same tbh. I can get Claude to spit out the first 80%, then I need to review it and hammer away at it for a time to get the last 80% :)