t23414321

Wouldn't 'thinking' need to be updating the model of reality (LLM is not yet that, just words) - at every step doing again all that extensive calculations as when/to creating/approximating that/better model (learning) ?

Expecting machines to think is.. like magical thinking (but they are good at calculations indeed).

I wish we didn't use the word intelligence in context of LLMs - shortly there is Essence and the rest.. is only slope - into all possible combinations of Markov Chains - may they have sense or not I don't see how part of some calculation could recognize it, or that to be possible from inside (of calculation, that doesn't even consider that).

Aside of artificial knowledge (out of senses, experience, context lengths.. - confabulating but not knowing that), I wish to see an intelligent knowledge - made in kind of semantic way - allowed to expand using not yet obvious (but existing - not random) connections. I wouldn't expect it to think (humans think, digitals calculate). But I would expect it to have a tendency to be coming closer (not further) in reflecting/modeling reality and expanding implications.