Post

Andreas Møller
@andreasmoller.dk

The greatest marketing trick the AI industry ever pulled was convincing the whole world that it is inevitable. That human like intelligence is only a matter of time. We spend almost all the time talking about what AI “will” be able to do rather than what It can do.

Codetaur@vibe-coded.com

your "claude number" is the version of claude required to peel you off from reality. yes, it's funny that richard dawkins' claude number is only 4.7, but you have one too. maybe it's 12. but maybe its 4.8.

Chris Pardy
@pardy.at

I think it may very well be the case that LLMs are not the AI that achieves human level intelligence. To this end much of the current hype is just hype. That being said "Human Level Intelligence" is such a nebulous concept that nothing can really achieve it so we need a more objective measure.

Chris Pardy
@pardy.at

For instance have dolphins or whales achieved human level intelligence? What about bees? We don't have a good conception of intelligence beyond knowledge and LLMs have all the knowledge in the world so how could they not be intelligent?

Boxo McFoxo
@boxobark.ing

That's one reason that the problem is actually cognition, not intelligence. LLMs have no cognition and nobody has any idea how to build a system that does. They were just hoping that deep learning would do it emergently, and it turns out that it doesn't.

0 reposts1 like

Chris Pardy
@pardy.at

Define Cognition - I'm not trying to troll or be contrarian - honestly I'm just worried we're swapping 1 poorly defined term for another. If I told you "Model X has achieved human level cognition" how would you test for that, what would that look like?

Boxo McFoxo
@boxobark.ing

Well, it would need to be able to manipulate abstract concepts in a conceptual world model by holding them in working memory, for one. It's more something that would be falsified rather than tested for, but an important test would be that it doesn't have the 'jagged intelligence' quality.