Post #2438477

2026-04-19 13:40 UTC

@tshirtman@mas.to to clarify (should've made this explicit in the post) I'm being pushed more in the autonomous agent direction. Can't speak too much to specifics but it's much closer to "just let it build" than as a super charged stack overflow replacement. As for open weights I appreciate the suggestion, but I think part of my point specifically centers around not using paid options at work and free options at home—it matters a bit less with LLMs since there aren't UIs and keybinds and workflows to relearn but it still is different. Regardless even if the model weights are openly available the hardware to run them on isn't :)

Replies (2)

@tshirtman@mas.to 2026-04-19 13:45

@jam@tilde.zone To be fair, I do agree with your point about de-skilling, and i see (as many people do) a similarity between what happened to skilled manufacturers when their work became industrialized and they became factory workers, and what's (tentatively) happening in software engineering (not just now, but in the last few decades), i'm not sure LLMs will achieve that, but i'm sure many hope it will. "Just let it build" will lead to bad results, at least at the moment, they need adult supervision.

Open ##2438478
@jam@tilde.zone 2026-04-19 13:45

@tshirtman@mas.to (I'm implicitly assuming you're not implying I should use a free model running on CPU with limited RAM at work, as that would be my only free option outside of work atm. Maybe you are suggesting that but I don't think it has any chance for the task at hand. Regardless, all a hypothetical given I don't think any model makes me enjoy doing things in my free time more than I do without)

Open ##2438479