The battling ai bit you mentioned. This is my life right. Ai is both amazing and shit. I feel like everyone else is running dark factories and producing millions of lines of code and having amazing lives. Meanwhile I am going insane with stress because I've burnt so much time trying to wrangle it on a team I've just joined. My productivity has not been good. I half feel like I am being gas lit by YouTubers and half feel "no I'm just doing it wrong"
I am not convinced that the "dark factory" / "gas town" people are actually shipping anything that isn't also part of the AI ecosystem. At least the noise/ship ratio is incredibly high.
The leading edge models are pretty good but we are still at the "rain man" stage so it's "jagged" intelligence.
It may be three years or so before the new compute-in-memory devices fully make it out of the lab and increase efficiency by about 100 times, allowing us to deploy models with human level complexity (100T vs current 10T SOTA) at scale.
> I feel like everyone else is running dark factories and producing millions of lines of code and having amazing lives
That is when I realize I'm spending too much time on HN. Because it is really only here that this vibe is so strong. My impression is that there is a lot of motivated reasoning in the folks that frequent HN.
It may also be that I work a boring job. If I turned up our code output by 10x it would not improve anything about our product. People who are pumping out dramatically more code have to be in an entirely different world. Or, you know, they're full of shit.
You are not crazy.
> I feel like everyone else is running dark factories and producing millions of lines of code
If they are, they aren't producing anything useful with it. Just look around - do you see a sudden increase in actually useful software alongside the AI boom?
What they are mostly doing is a snake-eating-it-own-tail million lines of code LLM harness to burn tokens faster to write more code... to write a 10 million lines of code LLM harness. Or endlessly bikeshedding the perfect LLM-powered bespoke personal knowledge base.
In normal software engineering jobs, we're debugging problems a bit quicker, we're writing boilerplate faster, we have a lot of questionable new test suites... but the game is more or less the same as it was before