Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by jppittma on Monday October 09, 2023 @03:32PM
(#37824371)
It's because up until a few years ago, everything that was capable of communicating at that level was a (non Artificial)GI. It's a bit ..surreal? to see text and speech without intelligence and understanding behind it.
[ Parent
| on HN ]
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by tsunamifury on Monday October 09, 2023 @04:04PM
(#37824795)
What is intelligence. Honestly other than something we keep claiming only humans have.
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by __loam on Monday October 09, 2023 @04:39PM
(#37825216)
You're making this comment in a way that devalues human intelligence but I actually think not having a great understanding of our own cognition is indicative of something deeper going on physically in our heads that isn't modeled by the extremely simple "neurons" in LLMs. I've seen several people here talk about "the weights in our heads" and I don't think we have enough information about the brain to make that analogy.
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by tsunamifury on Monday October 09, 2023 @08:49PM
(#37827350)
Did you know we also don’t know how LlMs work and neither do they…
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by jppittma on Monday October 09, 2023 @08:21PM
(#37827122)
What is the feeling of sunshine on your face? What is love? I think intelligence, as a prerequisite requires consciousness. It's the capability for the "aha!" moment where for a brief moment all if quiet, and the student and the concept are one.
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by tsunamifury on Monday October 09, 2023 @08:49PM
(#37827345)
This is ridiculous. Sorry. Intelligence is the ability to reason and communicate to another creature and for them to understand and cooperate or betray you.
This can do that.
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by Jensson on Tuesday October 10, 2023 @02:24AM
(#37829073)
A basic diplomacy AI is also intelligent according to your definition, so why do you think ChatGPT is different than what we had before then?
Re:LLMs can't self-correct in reasoning tasks, DeepMind stud...
by tsunamifury on Tuesday October 10, 2023 @09:05PM
(#37839790)
Yea we’ve had simple intelligence for a long time. What’s so controversial about that?