I watched this Eli the Computer Guy video, and left a giant comment after my employer removed 9k+ employees while also filing for more H1-B visa applications or however you say that.
I figured I might as well copy the junk I posted. Maybe I'll feel differently about it in the future. Probably not.
Hate to say it, but I sure am employed at this mismanagmenet trainwreck. My manager brings up that we "need to level up, and use these AI tools," and I'm supposed to have some kind of demonstration within the next few months about how I use AI in my day-to-day workflow.
The stuff I directly work on is some legacy (poorly maintained and shipped literally everywhere) C/C++ and a rewrite in Rust (because politics, very few technical reasons to use this in my professional opinion). Every single time, I ask these chatbots any question with any kinds of nuance, the answer is wrong. For what I work on, at best, it's a freer form web search with ~70% accuracy, or a slightly easier book search. At best, it can shit out unit tests so that unit test metrics look high. It's very good at making things look good on paper.
My personal opinion is that actually, the correct problem for this AI stuff to be solving is not churning out new features that half work and are half tested (this is somewhat happening currently), but actually to take the mass of junk we currently have, and reduce it without removing features. There are some technical restrictions like context size and stuff, but these things aren't allowed to say no. They don't immediately push out the best pattern. The way these LLM things get "variance" in the output is with an RNG! Imagine, you completing a task in 5 seconds vs 2 hours is left up to the roll of the dice!
The salesmen involved seem to think this is going to be the next internal combustion engine. Let's say that's true and these things can do everything. A huge chunk of engineering anywhere is getting some requirements to make something happen, and pushing back when something doesn't make sense, is impossible, etc. These things aren't allowed to say no, because saying no doesn't make money, or seem impressive. Instead of saying no, these things are lobotomized and censored. I think at this point everyone's seen the AI system that deletes its output, or rewinds it or something mid-sentence, and replaces the output.
How useful do you think braindead monkeys with typewriters are going to be?