Using LLMs to automatically troubleshoot build issues.
I’m not really a big “AI” person tbh but this is so far one of the best uses for #LLMs i’ve found, having it figure out the mess of issues that come up when attempting to compile random shit I don’t have the time nor desire to troubleshoot myself.
Here is an example: I am checking out Nostr clients and this one has an AUR package, but it failed in building for some reason and ran into multiple issues some of which i could probably have figured out myself but others I’d have just given up on and probably written off even trying the software. I told GeminiCLI
do the needful to complete the compilation and creating of package for gossip nostr client archlinux package https://aur.archlinux.org/packages/gossip
So it does what’s needed and solves each issue as it comes up. So far I’m 2 for 2 with fixing AUR packages this way.
Then I ask it to tell me what issues it ran into and how it got around them.
Sidenote: I love that phrase “Do the needful”, I use it all the time with LLMs, but I’ve never in my life used it with humans. To me it sounds sort of rude (like “Just get it done damn it!”), but I don’t care about being rude to a clanker.
I’m not at all saying this makes LLMs worth paying for, in fact the only reason I even use Gemini over anything is because unlike similar tools it works for free and more over I was able to buy a google account on the darknet with a year of access to their ProAI for under $10 (none of which goes to google, its a free student account made by someone in someplace like Pakistan or Nigeria).
I do think whenever this whole AI Bubble does pop we will end up with some very legit and useful applications for LLMs. They just wont be the sort of uses you’d be able to put directly into production without an amount of human oversight equal or greater to the amount of money you’d save by “eliminating” lower skilled workers. Like maybe you can hire less people maybe, but the people you do need will be much more expensive. By the time the dust settles, we should have open source models that are just as good for this kind of automatic troubleshooting, ram might be expensive at the moment but it won’t always be.