@tante this really depends on how you define intelligence.

if intelligence a priori requires consciousness then yeah llms will never have that

if intelligence is defined by capabilities like doing something by solving some problem and is scene as not one thing but many things then an llm is closer to human levels of capabilities than any prior expert system and fails more like a person as well.

ned block calls them zombies but i think how we train them is more like ghosts/shades of the collective subconscious + whatever the models makers biased into it

i dont like this but also not surprised we can simulate human intellegence just at 1000x the energy cost and a 1/10th the longeviety of hardware. simulation isnt the same as implementing though which is why its so inefficient. i.e. its a blockhead https://en.wikipedia.org/wiki/Blockhead_(thought_experiment)

but a compressed and loosy version that makes up what it doesnt know because a complete and correct blockhead is the size of the universe just to exist on earth will human senses