Let’s put aside for a moment all the ethical/ecological/societal considerations […]
I sometimes feel we’re putting aside too many factors and focus on some hypothetical debate which doesn’t reflect reality.
I mean my point would be, you’ll immediately accept low-quality contributions. And those coding agents for some reason love to generously rewrite stuff and touch a million different files. So what I’ve seen is, you almost immediately end up in some maintainability nightmare which you might or might not have the capacity to deal with (properly). The broad idea by the AI companies is to then use AI to deal with AI. But I’m not sure if that’s a great idea.
I find that it stinks that we let get companies away with simply discarding ethics and existing laws.
Look, I can’t go to a party and tell people, “well, I am a robber, my day job is to mug old ladies at ATM machines and stabbing them to death, then take their money. But this is just my job, it is the economy, bro, we cannot let ancient moral rules interfere with the realities of business!”.
Why are large corporations allowed to behave like this? Like standing above the law?
Good question. I feel we here in Europe have a bit more regulation regarding companies in general. But for some reason the big tech companies get away with almost anything. They can do arbitrary evil business models. Own everybody’s private data. Do harm to society. They don’t have to pay taxes. And now we got AI and they’re allowed to steal as well. And some more.
Im Kapitalismus herrscht das Kapital.
Who do you think is making the laws and deciding which ones are enforced? Capitalists.
There’s a lot of material about this. By chance I came across this wired history video that’s kinda related https://youtu.be/4GRVWIBYvOM
I switched teams at $DAYJOB a few months ago, into a project which is developed open-source mainly by our company and with heavy use of LLMs.
And at some point, one of our devs told me that they explicitly don’t have an
AGENTS.mdfile in the repo (which would contain instructions to make the LLMs generate better results), because they don’t want LLM-generated contributions.Why? Because we would need to review that slop.
I feel like this isn’t talked about enough in these discussions. It’s faster to generate a contribution, but the end result is often considerably worse and needs a much more thorough review.
The workload for maintainers increases. And maintainers were the bottleneck even before LLMs.So, I just don’t think you generally want to allow generated contributions, not without LLMs making a massive leap still.
At best, I would allow it for select people, like the maintainers themselves or past contributors where you know they will ensure a base level of quality and have an understanding of the architecture and the project goals.This is the wrong question to be asking. “Is the code good” is literally all that matters when writing software that others will use. Slop is slop no matter if it comes from a person or an LLM. A human should be reviewing all code.
The problem is “slop code” and “bad code” are always the same thing. That’s why it’s called slop code.
Agreed!
“Is the code good” is literally all that matters.
Only if one ignores the ethical, environmental, and copyright concerns.
How dare you have nuance!
Those are imaginary shit made up to stop progress. AI will be used. If not then the project will be formed and the fork will get leaves better than the human slop version. Meanwhile you’ll keep coping that le AI bad with le enbyornmint. AI uses less water than growing almonds. Stop being a brainwashed cattle
AI will be used. If not then the project will be formed and the fork will get leaves better than the human slop version.
I heavily doubt this is actually the case.
But perhaps you can help out and show me some convincing real-world examples where large, mature, abd widely used FOSS projects were forked and became significantly better because use of LLM code generation?
We have these things now for at least a year and as the old saying from the dot com era goes, Internet years are dog years, meaning a three year old dog is adult, so there must be a really successful application of that LLM technology in software development?
Thing is, you can eat and process almonds further, ai slop? Not so much.
AI uses less water than growing almonds. Stop being a brainwashed cattle
Citation needed.
The issue with the almonds is that there are a bunch of dumbasses growing them in an area that, left to itself, would be a desert. They shouldn’t be doing that, either.
Irrelevant whataboutism. Nothing about the almonds justifies or is at all related the datacenter build-outs.
My point was that even if the water use of data centers is less than that of almonds, it does not justify the existence of the data centers because the water use of the almonds is also grossly inflated. You don’t waste 999 million dollars just because someone else is wasting a billion.
If you want to survive? Yeah. Besides AI isn’t a ego driven maniac like some devs so it’s a bonus.
I almost never use LLM code except as last resort. So, no. And less than .1% of my code are LLM. And I do have to verify, and I rather verify my own anyway.
is this small part worth the effort?
