Hacker News Reader

Comment by fabrice_d | Hacker News Reader

And still, Yann LeCun (head of AI at Meta, renowned AI/ML researcher) is convinced that we are far from reaching AGI. He makes convincing arguments, especially around the fact that we are not able to expose models to the amount of redundant information that even a young kid is exposed to.

I guess we'll see some shifting goals around "what is even AGI". You say the we'll have soon "models that most of us consider AGI today", but what does that even mean?

View on HN

The point from Yann LeCun I find interesting is the negative space argument where as training data / parameters increase, the "best" next tokens represent a smaller and smaller slice of the model. His contention is therefore more hallucinations, more places to get stuck on some less best next tokens, etc and interesting to think about this as the opposite of how scaling laws are typically presented. A lot of smart people stabbing around in the dark right now and only time (and gazillions in GPUs) will tell.

I mean, this new openai model is trained on video and voice as well.

So already, his argument is outdated. He had made a few of these claims over the last year that have been quickly disproven by a new model.

They are trained on a ridiculously small amount of content compared to what the brain of a 3 years child is subject to. Current models still have a very narrow application field compared to what a human can achieve.

Reply on news.ycombinator.com