‘But there’s probability!’ is such a weird nitpick, when any randomness is obviously added on purpose. You can just… not. It’s literally matrix algebra. Noise is a generally useful tweak for a glorified chatbot.
Good enough to be a problem for open-source, since it can convert a whole project to another language, or just rewrite everything in a different structure.
If it simply Did Not Work, there’d be nothing to discuss. But the unavoidable truth is that chatbots can code, now, and we all have to deal with it. If randomness is more trouble than it’s worth, well then hey guess what, determinism is the default.
The python library chardet was recently recreated with all-new code, with a completely different structure and mechanism from the prior version, entirely by an LLM. It’s causing a huge fight in the community because the-- author? People get real snippy about language here. The guy who ran the LLM, incidentally also the only reliable maintainer of chardet for the last decade, said the new version is distinct and can’t be copyrighted so it’s BSD-licensed.
Minecraft seed-hunting ultranerds have thrown their esoteric cryptographic Java code into Claude, said “rewrite this as Rust,” and seen massive performance improvements.
I have seen several machining and robotics Youtubers hand-wave their way out of integration hell by letting the machine figure it out.
There’s a lot of damning things to say about LLMs and LLM coding, but a flat ‘it doesn’t work’ is wishful thinking.
I really wish we had a thing that does what I’m told it does. I think that would be cool. The weird fanatical pretending and cherry picking really gets to me. So much worse than it was with the equally useful nft’s
Even if these real examples were cherry-picked - they’re still real examples. In these cases, it did the thing. That success will be pursued, expanded, and streamlined, so it is more likely to do the thing. Comparisons to an outright scam are nonsense because this demonstrably functions.
It does not prove merit. Sorry. I don’t trust any ‘accomplishment’ of these systems in a field I do not understand thoroughly after being told about all their accomplishments in fields I did understand thoroughly that were bullshit pushed by people who only vaguely kind of understood what they were saying and were mostly wrong. Too many bullshit liars and cherry picking fanatics who understand nothing. Poisoned well.
‘But there’s probability!’ is such a weird nitpick, when any randomness is obviously added on purpose. You can just… not. It’s literally matrix algebra. Noise is a generally useful tweak for a glorified chatbot.
Not even a good chatbot
Good enough.
Good enough to bicker about maintainability.
Good enough to be a problem for open-source, since it can convert a whole project to another language, or just rewrite everything in a different structure.
If it simply Did Not Work, there’d be nothing to discuss. But the unavoidable truth is that chatbots can code, now, and we all have to deal with it. If randomness is more trouble than it’s worth, well then hey guess what, determinism is the default.
No they can pretend to code.
Chatbots can bullshit. They aren’t really good for anything.
The python library chardet was recently recreated with all-new code, with a completely different structure and mechanism from the prior version, entirely by an LLM. It’s causing a huge fight in the community because the-- author? People get real snippy about language here. The guy who ran the LLM, incidentally also the only reliable maintainer of chardet for the last decade, said the new version is distinct and can’t be copyrighted so it’s BSD-licensed.
Minecraft seed-hunting ultranerds have thrown their esoteric cryptographic Java code into Claude, said “rewrite this as Rust,” and seen massive performance improvements.
I have seen several machining and robotics Youtubers hand-wave their way out of integration hell by letting the machine figure it out.
There’s a lot of damning things to say about LLMs and LLM coding, but a flat ‘it doesn’t work’ is wishful thinking.
I really wish we had a thing that does what I’m told it does. I think that would be cool. The weird fanatical pretending and cherry picking really gets to me. So much worse than it was with the equally useful nft’s
Fingers in ears is not an argument.
Even if these real examples were cherry-picked - they’re still real examples. In these cases, it did the thing. That success will be pursued, expanded, and streamlined, so it is more likely to do the thing. Comparisons to an outright scam are nonsense because this demonstrably functions.
It does not prove merit. Sorry. I don’t trust any ‘accomplishment’ of these systems in a field I do not understand thoroughly after being told about all their accomplishments in fields I did understand thoroughly that were bullshit pushed by people who only vaguely kind of understood what they were saying and were mostly wrong. Too many bullshit liars and cherry picking fanatics who understand nothing. Poisoned well.
Facts aren’t real because scams exist. Sure, dude.