>>21972694>I'd argue that what we already have is able to reason at novel logic problems already, just not quite at the superhuman level you describeThey are not, and they utterly fail when you give them simple but novel problems to solve that even most children would easily be able to solve. They don't even know how to count. Pic very related.
>So far there aren't really any issue with scalingOf course there is. They've run out of data to train the models on. They're paying millions of dollars to randos all of the internet to generate novel training data, but it doesn't seem to be doing much. The models are already good at common tasks/topics, and there simply isn't enough data to make them any good at niche topics or rare types of problem solving.
>and with the rate of development we've been seeing in the past three monthsThere has been no significant improvement over the past three months. You're buying into the hype.
