I've been part of open source for most of my life. This year, the projects I love started closing their doors to outside contributors, and I don't blame them one bit, but it makes me sad.
A personal look at how AI is changing open source as I've known it.
I’m not saying they’re awesome. I’m agreeing with the grandparent that says they are:
awesome for end users
You can dislike LLMs because they suck the fun out of programming, but some people don’t like programming so it doesn’t make sense to dislike how other people use them to avoid that!
Even the most LLM-happy programmer I know is clear that you need to verify the LLM’s work. Does the end user who doesn’t like programming suffer through that verification? Or do they just use the output unchecked, which I assume you’d agree is not “awesome”?
The real line is not whether they can check the output. The real line is whether they can both understand and accept the error rate.
And there are not so few things where (a) «works 90% of the time, gives weird garbage the remaining 10% of the time» is useful if it is cheap enough, (b) getting to 90% does not take too many «Please fix this problem» promtps (with manual retesting of previous issues, as by assumption forcing the LLM to build a reliable regression test suite is outside the user’s qualification).
To borrow the example elsewhere from the conversation on this page — sometimes one needs to copy a playlist perfectly because this is the music for a carefully staged event; sometimes even a maliciously selected 90% out of a shuffled playlist is close enough for a morning run.
Have you seen anything Web done by a large corporation in the last 15 years? This is already how software looks like outside of the best-of-the-best command-line tools (so all software the user from the hypothetical uses is not much better).
I have, and it’s indeed almost always terrible. Every minute that I use Teams I curse the developers who made it and the management who approved it, and their lineage for 7 generations, and I wish the earth they’ve stood on salted!
In life in general I’d like more good stuff and less bad stuff. So a machine that lets most people push out the terrible quality you describe is a bad thing (not a bad thing in the sense “we should make it illegal and stop it”, but in the sense “we should all frown and admonish them, our human peers, sternly for using this shit”).
It’s like a temu factory in everyone’s pocket. Decidedly not “awesome”.
Pretty sure that users that persist get results better than Teams. It takes effort and qualification and five layers of management to build something that bad.
What they get is typically better than the average of what they get otherwise, and fills a gap where they won’t get anything at all otherwise. And importantly, they do not push this out, they use it themselves, if they manage to convince the mad alien daemon to get it into the shape where using this is more comfortable than not using it. Probably more durable than temu stuff, too, would definitely be more durable if we could regulate breaking changes in OSes.
This is better than the options they realistically had otherwise. If you count it as bad stuff, by that logic your argument also applies for bringing down technological civilisation, which would probably imply gigadeaths, but hey, it would imply less stuff and most of stuff is somewhat bad, so.
All this is not an argument to oppose anything that goes against hyperscale tech obviously, building non-technical-user-oriented UI around a harness for a medium-size (20B–120B) model is an investment nobody wants to make against the oligopoly, but without oligopoly it will go the way of imagegen UI-wise which would be good enough, and medium-size LLMs are probably good enough as is (and unlike 1T-scale, non-oligopoly actors also create them).
I don’t have a say in how other people use their computers, but that doesn’t disqualify me from making a judgement about whether I think what they’re doing is a good idea or not.
You yourself say what they do is “awesome”, so you’re clearly allowed an opinion. Are only positive ones acceptable?
I’m not saying they’re awesome. I’m agreeing with the grandparent that says they are:
You can dislike LLMs because they suck the fun out of programming, but some people don’t like programming so it doesn’t make sense to dislike how other people use them to avoid that!
score 1 · canonical
Even the most LLM-happy programmer I know is clear that you need to verify the LLM’s work. Does the end user who doesn’t like programming suffer through that verification? Or do they just use the output unchecked, which I assume you’d agree is not “awesome”?
score 1 · canonical
The real line is not whether they can check the output. The real line is whether they can both understand and accept the error rate.
And there are not so few things where (a) «works 90% of the time, gives weird garbage the remaining 10% of the time» is useful if it is cheap enough, (b) getting to 90% does not take too many «Please fix this problem» promtps (with manual retesting of previous issues, as by assumption forcing the LLM to build a reliable regression test suite is outside the user’s qualification).
To borrow the example elsewhere from the conversation on this page — sometimes one needs to copy a playlist perfectly because this is the music for a carefully staged event; sometimes even a maliciously selected 90% out of a shuffled playlist is close enough for a morning run.
score 1 · canonical
This sounds like all the worst parts about the non-software world brought to the software world. Yuck!
score 1 · canonical
Have you seen anything Web done by a large corporation in the last 15 years? This is already how software looks like outside of the best-of-the-best command-line tools (so all software the user from the hypothetical uses is not much better).
score 1 · canonical
I have, and it’s indeed almost always terrible. Every minute that I use Teams I curse the developers who made it and the management who approved it, and their lineage for 7 generations, and I wish the earth they’ve stood on salted!
In life in general I’d like more good stuff and less bad stuff. So a machine that lets most people push out the terrible quality you describe is a bad thing (not a bad thing in the sense “we should make it illegal and stop it”, but in the sense “we should all frown and admonish them, our human peers, sternly for using this shit”).
It’s like a temu factory in everyone’s pocket. Decidedly not “awesome”.
God help us all.
score 1 · canonical
Pretty sure that users that persist get results better than Teams. It takes effort and qualification and five layers of management to build something that bad.
What they get is typically better than the average of what they get otherwise, and fills a gap where they won’t get anything at all otherwise. And importantly, they do not push this out, they use it themselves, if they manage to convince the mad alien daemon to get it into the shape where using this is more comfortable than not using it. Probably more durable than temu stuff, too, would definitely be more durable if we could regulate breaking changes in OSes.
This is better than the options they realistically had otherwise. If you count it as bad stuff, by that logic your argument also applies for bringing down technological civilisation, which would probably imply gigadeaths, but hey, it would imply less stuff and most of stuff is somewhat bad, so.
All this is not an argument to oppose anything that goes against hyperscale tech obviously, building non-technical-user-oriented UI around a harness for a medium-size (20B–120B) model is an investment nobody wants to make against the oligopoly, but without oligopoly it will go the way of imagegen UI-wise which would be good enough, and medium-size LLMs are probably good enough as is (and unlike 1T-scale, non-oligopoly actors also create them).
score 1 · canonical
I can’t imagine anything I’d care about less as if it’s for their own personal use.
score 1 · canonical
Then we’re just very different people.
score 1 · canonical
Why do you care how people use their computers? And how does it have anything to do with your original argument that it takes the fun out of things?
score 1 · canonical
I don’t have a say in how other people use their computers, but that doesn’t disqualify me from making a judgement about whether I think what they’re doing is a good idea or not.
You yourself say what they do is “awesome”, so you’re clearly allowed an opinion. Are only positive ones acceptable?
score 2 · canonical