@jordgubben@mastodon.gamedev.place
Post #1295866
2025-07-26 15:51 UTC
@jigso @jk As I’ve come to understand it, it’s only considered ”AI” if it’s mostly useless. We don’t consider the statistical models in auto complete (or spell check) as ”AI” because it kind of works fine. Same with facial recognition in photo album software, or navigation in robotic vacuum cleaners.
Thus, since the rubber duck works it can not be considered fulfilling this criteria for ”AI”.
Replies (1)
-
@yacc143@mastodon.social 2025-07-27 16:27
@jordgubben But it is not useless. You can use LLM for useful stuff. But telling a LLM to write you a pink pony does not work. (Literally from a company internal presentation.) Breaking the job down into multiple steps (doing the architecture by hand), and then using a LLM to process the text. And extract data from text. And that worked surprisingly well. Now most AI assistants for coding are rather primitive on the algorithm side how they deal with the LLM. Simple example @jigso @jk