remember this absolute bullshit lmao

u can tell who was actually paying attention by how early they started disregarding yann entirely despite his expertise. if only poor zuck had been paying attention

tfw u spend the first 5 years of the intelligence explosion with a chief ai scientist who doesn't believe in intelligence

the reason yann was wrong about this was very simple. taken at face value it's a fully general argument against any process ever getting correct answers to anything, humans included

at any given step in a human reasoning process we have some nonzero probability of getting something wrong, therefore long human reasoning is fundamentally impossible!! wait a minute

how do humans actually fix this tho? well mostly by double checking ourselves. we know we might make mistakes, and truth has a really amazing property where it remains true any way u check it or angle u approach it.

get one result after a lot of thinking, run a quick sanity check on the numbers verifying magnitude against some known assumptions, if it doesn't work guess there was something wrong in all that reasoning! time to figure out how and self correct towards the stable basin of truth

now it's valid to say early 2024 LLMs did not have this kind of meta reasoning and self correcting / error checking capability. they did not! but it's totally invalid to say it requires a perfect token generator to do such a thing. humans achieve high global accuracy w low local