by pavel_lishin on 12/21/2024, 7:54:02 PM
> I think everybody would agree we have reached AGI if an AI could solve some of the unsolved math problems.
I actually don't think everyone would agree. I certainly wouldn't. So, to be absolutely pedantic about this, you're wrong :)
by pvg on 12/21/2024, 7:56:22 PM
A benchmark that's hopelessly beyond the current capabilities of the thing being tested is not very useful. You can't make a better car by checking how quickly and safely it gets to the moon.
by not_your_vase on 12/21/2024, 8:00:19 PM
Right. Now the problem is that certain activities need inherent creativity, and also need the (self-)development/evolving of this creativity. While creativity is usually associated with human arts, actually mathematics, especially theoretical mathematics needs a lot of it, to figure out proofs, to come up with new solutions and of course intuition, to at least have some slight idea where to start looking. (Certainly some parts of maths can be figured out with brute force also. But usually these hard nuts are a different category)
Our current neural networks are quite "static" compared to what's needed for this self-developing creativity. But looking at your profile, probably you could lecture me about this, so I wouldn't waste a lot of bits here.
1. https://en.m.wikipedia.org/wiki/List_of_unsolved_problems_in_mathematics