you are viewing a single comment's thread
view the rest of the comments
[–] 2 points 20 hours ago*

Well put. My day job is training AI for major corporations and governments and my point has become the same: its an advanced autocorrect feature, built on a model and architecture that guesses the best answers based on language patterns and nothing more.

For programmers to use them to remind themselves of the fastest way to iterate a list, or map a matrix, it is amazing. But anything more than a single function in size and its immediately risky for the same reasons it is a good guesser.

Programmers have turned out to spend MORE time coding than before LLMs https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/, from my experience that is due to not understanding the limitations of the architecture. But that so means admitting they aren't that game changing as we were sold.

Especially with art. LLMs are absolutely dog shit at art if you need to iterate for accuracy.

  • source
  • parent