„Wie Alien Intelligence“: KI liefert korrekte Ergebnisse durch Berechnungen, die die Wissenschaft noch nicht erklären kann
https://www.fr.de/wissen/wir-haben-keine-ahnung-welchen-effekt-die-ki-da-erzeugt-hat-forscher-stehen-vor-ki-raetsel-94033609.html
4 Kommentare
The article discusses a striking problem in current AI research: certain models are producing correct results through processes that remain unclear or impossible to interpret. If this trend continues, how do we evaluate, verify, or responsibly use systems whose internal reasoning we cannot fully understand? What does this imply for scientific progress and future oversight of increasingly opaque AI models?
The question I guess is how does it then create the nonsense that it is if anything more famous for? Have we just created a very new Magic 8-Ball that requires a data centre?
Don’t get me wrong, I like AI tech, what I’ve seen so far a lot of it is very impressive. But if you can’t rely on it, if you don’t know how it is doing what it does, then that doesn’t feel reliable. And if it’s not reliable then you have to massively reduce the number of potential uses. A technology that is wrong one time in a hundred is no good for doing tasks at scale, or when the stakes are high.
All LLMs at the moment are trying to (statistically) imitate the monkey without having a monkey to start with. Some may call it Alien most will call it garbage.
Even a broken clock is right twice a day. Doesn’t mean it’s worked out a new method of timekeeping.