Forgive me, I wrote a thing about AI:

[munificent@mastodon.social]

Forgive me, I wrote a thing about AI:

https://journal.stuffwithstuff.com/2026/01/24/the-value-of-things/

[sarna@types.pl]

@munificent I agree with a lot that you wrote, especially about the time & effort's relation to meaning. however, I think you're being too charitable to AI in this post. the assumption is that the utility is there - that using AI is actually faster, or you can get a better result in the same time. or that the generated code has more value (err, utility) than mental models in people's heads. I don't see it in the industry at large (where are all the new apps? [1]), and I don't see it in studies [2] (at least the ones not sponsored by microsoft/openai/whatever). I get it that it's smart to assume that The New Model will solve everything in a few weeks. as it's always claimed, but.. do you really want to believe in that?

1: https://mikelovesrobots.substack.com/p/wheres-the-shovelware-why-ai-coding
2: https://arxiv.org/abs/2507.09089

[munificent@mastodon.social]

@sarna

> the assumption is that the utility is there

The article does state that assumption explicitly, but yes it's an important point. What I'm trying to work through is that to the degree that AI does do useful stuff, when would I want to use it or avoid it?

[vz@fosstodon.org]

@munificent Under this assumption, the answer to this question seems to be quite simple: you'd use it when only achieving the result counts.

And there certainly are circumstances when this is the case, so the really unclear thing is what kind of things LLMs can genuinely help with. E.g. if they can really produce novel medicaments at a fraction of the time and price it currently takes to do it, they would certainly be worth the hype, wouldn't they?

[munificent@mastodon.social]

@VZ

> they would certainly be worth the hype, wouldn't they?

Possibly, but there is still the question of externalities and societal impact. Internal combustion engines are excellent for helping get people where they want to go quickly, but that doesn't mean they are overall worth it to the world.

[isntitvacant@hachyderm.io]

@munificent well put!

(another element of meaning, other than time, might be “vulnerability”: in the sense that all craft is rehearsal for later work, making something for someone else gives them a marker of where you were and how you thought about that craft. Which can be scary! You put a piece of a version of yourself into the work— as you were when you made that work — and you trust the receiver will value that piece, just as it reflects how you value them.)

[munificent@mastodon.social]

@isntitvacant Oh that's such a good point.

[vz@fosstodon.org]

@munificent This is not a popular viewpoint round here, but I think the sudden concern about externalities is weird. Nobody cared about video cards energy consumption as long as they were used only for gaming.

Societal impact is potentially much more important, but it really depends on what exactly are LLMs capable of. If they can do everything humans can do (which seems hard to believe, but then it's hard to believe they work as well as they do in the first place), then we're all hosed anyhow.

[andrewrk@mastodon.social]

@munificent I found it insightful. Thank you

[skybrian@mastodon.social]

@munificent I'm collecting metaphors about AI and one I kind of like is that, when you're using it to clean up your codebase and improve code health, it's sort of like using a pressure washer. You can use it to clean your steps but you wouldn't use it to clean a painting.

A lot of the code we write is not that delicate and can stand up to pressure washing, if it has good tests and all that.

I can take satisfaction in getting things nice and clean even when it's not hugely meaningful art.