I am saddened the emergent "fully-agentic" meta that wants to reduce the human programmer to a purely verification role. I do not want to be a QA tester for an LLM. I want to think/understand/control the code myself, even if I no longer write every line by hand. I want to retain my own agency (no pun intended).
This is all so depressing. I still use AI as a pair-programmer. I tell it what to code, and I hammer it until it gives me exactly what I want. So I am stuck in 2024-2025, I guess.
Though, I have been a lot more liberal with code I do not care as much about. I have mostly stopped making comments on my coworker's PRs... The code their AI writes looks unfamiliar, but there are no obvious errors, and it feels pointless blocking a PR over minor issues (ex: "hey this fn already exists in utils module", "hey, we do not annotate fn returns, because type checker infers automatically", "hey, you abstracted 2 lines of code into its own fn, but you only use it one place, so just inline it"). Haha, at least, I am not standing in the way of others, even if I read my own code still.
I desperately hope that eventually the industry will settle into some sort of a happy compromise, where human programmers can still have fun and express skill.
I think ai's special power is being convincing. Whatever it produces comes across as very persuasive and looks really good. The problems are mostly only noticeable if you dig in a bit and read between the lines or check assumptions etc. Then you might find out that the entire approach or paradigm being used is irrelevant and counterproductive... oops
In programming you've got a lot of decisions to take, many aren't verifiable and even more have tradeoffs so you can't hillclimb your way out of architecture design and requirements.
Most of what people call AI slop is literally making AI do design decisions. AI is much faster at executing than you are but making wrong decisions 20x as fast doesn't get you to fulfilling your requirements any faster.
From another search, it looks like go leapfrogged from ai-sucks to ai-is-unbeatable and didn't really get a centaur period.
Please correct me if wrong. I'm disappointed to learn this.
Can anybody clarify the relationship between that adversarial deep learning where it can play games against itself to improve and llms? I figure the clear win conditions in chess and go make it easier to train models, and I'm not very clear on how llms know if they improved or not