I watched @ylecun episode on @lexfridman and I think his bigger point is we should use a different architecture to actually make AI understand concepts Instead of just autocompleting word tokens like in LLMs or pixels/patches in image/video gen He wants an AI to actually understand the world, not just simulate it So maybe he's just one step ahead and thinks we're investing in the wrong type of AI which is just a simulation of intelligence instead of actual intelligence I'm not sure if I'm sure that's what he means though because I don't understand most AI terminology/jargon