Post #2111314
2026-05-04 09:57 UTC
#AI haters, your time to shine!
In a few weeks I'll probably have to make a presentation on full-cycle agentic development, and it's going to be a tough one. I'd appreciate links to any high-quality analysis of this topic specifically. Downsides, gotchas, misconceptions, you know. The more thoughtful and nuanced the better.
Note that I'm not looking for general AI criticism - I can produce enough of it myself. And no, the target audience is not much interested in purely ethical aspects, alas.
Replies (2)
-
@creepy_owlet@mastodon.online 2026-05-04 09:59
An example. There was a recent article proving that AI, if let to process its own outputs, cannot reach superintelligence. Quite the opposite. And that was proven mathematically (or so I understood, I'm not an actual smart person). Has anyone put thoughts into how it applies on smaller scale, i.e. on agents being left more or less alone on software, with only design guidance from humans? Can it be seen as a proven fact that the code will turn into chaos? Or is it likely but who knows?
-
@dogfox@kpop.social 2026-05-04 11:52
Here is a nice paper spelling out what should be obvious, that LLMs are not thinking: https://arxiv.org/html/2512.19466v1 @creepy_owlet