[ben_w]

> Speak for yourself. LLMs are a feedforward algorithm inferring static weights to create a tokenized response string.

If we're OK with descriptions so lossy that they fit in a sentence, we also understand the human brain:

A electrochemical network with external inputs and some feedback loops, pumping ions around to trigger voltage cascades to create muscle contractions as outputs.

[bigyabai]

Yes. As long as we're confident in our definitions, that makes the questions easy. Is that the same as a feedforward algorithm inferring static weights to create a tokenized response string? Do you necessarily need an electrochemical network with external stimuli and feedback to generate legible text?

No. The answer is already solved; AI is not a brain, we can prove this by characteristically defining them both and using heuristic reasoning.

[ben_w]

> The answer is already solved; AI is not a brain, we can prove this by characteristically defining them both and using heuristic reasoning.

That "can" should be "could", else it presumes too much.

For both human brains and surprisingly small ANNs, far smaller than LLMs, humanity collectively does not yet know the defining characteristics of the aspects we care about.

I mean, humanity don't agree with itself what any of the three initials of AGI mean, there's 40 definitions of the word "consciousness", there are arguments about if there is either exactly one or many independent G-factors in human IQ scores, and also if those scores mean anything beyond correlating with school grades, and human nerodivergence covers various real states of existance that many of us find incomprehensible (sonetimes mutually, see e.g. most discussions where aphantasia comes up).

The main reason I expect little from an AI is that we don't know what we're doing. The main reason I can't just assume the least is because neither did evolution when we popped out.