It's been years and I still don't understand why even FOSS projects are embracing AI. I just can't understand it.
My only guess is that since they are unable to figure out what code is produced with AI and what code is not, they'd rather make guidelines about it than trying to prevent what they might be unable to prevent.
But even then, a lot of them allow code review with it, which is INSANE. So basically you are handing your code review to a machine that hallucinates. Seriously? And I don't care how many of y'all swarm to me now telling me how "good" it is for review and how "AI is going to get better" when it's not. You know who you are...
I know Linus used the argument of "static analyzers are AI too and they also have red flags". Yes, but it's not an LLM even if it includes a semantic analyzer that tokenizes the code. It builds a tree of the code and matches patterns on that tree, looking for whatever it has to look for. You can read more about LLVM's static analyzer here: https://arxiv.org/pdf/2408.05657
Static analyzers are not perfect even without LLMs, which is why I think it's even worse if you mix an LLM into the equation. I know it looks like it works, I know you wish it worked, but it doesn't because it's very unreliable. Proof of that is that Microsoft multiplied the number of bugs in their patches and updates since they started using AI to code. And while AI has not been identified officially as the source of all problems, the timing is too perfect for it to be a coincidence. If it doesn't work for Microsoft, it doesn't work for anyone.
ENIGMATICO :flag_bisexual: :flag_nonbinary:
npub1qe…4nq5j
2026-03-07 20:33:00 UTC
Author Public Key
npub1qekcg7pwryemfla9eeek4pa2tpv4n4rlu0ncm0napxsmdmedk68q04nq5jSeen on
wss://relay.momostr.pinkPublished at
2026-03-07 20:33:00 UTCEvent JSON
{
"id": "1c2c1413d5266f51069979f2f6a25a65532a72e9f39a3fd47adf64a02ee497b3",
"pubkey": "066d84782e1933b4ffa5ce736a87aa585959d47fe3e78dbe7d09a1b6ef2db68e",
"created_at": 1772915580,
"kind": 1,
"tags": [
[
"proxy",
"https://mk.absturztau.be/notes/ajkch27b286p03oo",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://mk.absturztau.be/notes/ajkch27b286p03oo",
"pink.momostr"
],
[
"-"
]
],
"content": "It's been years and I still don't understand why even FOSS projects are embracing AI. I just can't understand it.\n\nMy only guess is that since they are unable to figure out what code is produced with AI and what code is not, they'd rather make guidelines about it than trying to prevent what they might be unable to prevent.\n\nBut even then, a lot of them allow code review with it, which is INSANE. So basically you are handing your code review to a machine that hallucinates. Seriously? And I don't care how many of y'all swarm to me now telling me how \"good\" it is for review and how \"AI is going to get better\" when it's not. You know who you are...\n\nI know Linus used the argument of \"static analyzers are AI too and they also have red flags\". Yes, but it's not an LLM even if it includes a semantic analyzer that tokenizes the code. It builds a tree of the code and matches patterns on that tree, looking for whatever it has to look for. You can read more about LLVM's static analyzer here: https://arxiv.org/pdf/2408.05657\n\nStatic analyzers are not perfect even without LLMs, which is why I think it's even worse if you mix an LLM into the equation. I know it looks like it works, I know you wish it worked, but it doesn't because it's very unreliable. Proof of that is that Microsoft multiplied the number of bugs in their patches and updates since they started using AI to code. And while AI has not been identified officially as the source of all problems, the timing is too perfect for it to be a coincidence. If it doesn't work for Microsoft, it doesn't work for anyone.",
"sig": "d6dc9de403351b1d6562c674dd3ea21c6763e770e62113bc28487ccc32c9ac1792bef35a96e6a698b0c1f5c9f96c88b3ad3d8f2ec08102c167562812b331749b"
}