Oh yeah, 100% agree the future is local. I think we’ll have dedicated chips that have specific models burned on to them to run ultra fast and efficiently.

The mainframe style of computing always goes out of vogue as soon as it can because it sucks

sure - but the unpaid volunteer building your free and open source software wants to go faster so they can spend less of their valuable time on it.

In-general, if you feel this way, lead by example. Fork or contribute but don’t just complain

So do I. You can structure and use the tools responsibly.

Two things I really like using LLMs for right now:

1). Search complex codebases for summaries of “how does this work”. Especially if you are working outside of a project you normally work in, but your code still utilizes it and you want to understand some behavior (or at least know where to look).

2.) PR reviews. I’ve been building a custom skill for awhile now that does a great second pass on PRs. I do my initial review, then sic the LLM on it. It often turns up small things I overlooked that are worth addressing.

Currently, I use LLMs in more of a read-only manner, but I have had success in giving them well-structured easier tickets, if your project has good guidelines and you use the planning mode. You need to have an understanding of where you are working to even utilize these.

I know it’s an unpopular take, because the hive mind wants LLMs to fail SO bad here, but I think there is a usecase for these long-term for B2B software dev.

That said, I generally disagree with the shoving LLMs into things. There are a lot of wasteful examples where companies replace a perfectly good deterministic thing with a token generator and then it gets worse.

yes - God forbid people try to use tools to enable them to get things done faster. If folks worked in software they’d see that LLMs will not be going away there. Folks need to understand that FOSS is not an exception here. They’re welcome to fork and maintain things thanklessly themselves if they dislike it.