DeepMind AlphaGeometry: Is AGI just around the corner?

  • Thread starter Thread starter Meow12
  • Start date Start date
  • Tags Tags
    Ai Mathematics
AI Thread Summary
DeepMind's AlphaGeometry demonstrates impressive capabilities in solving Euclidean plane geometry problems, performing nearly at the level of human gold medalists in the International Mathematics Olympiad. This achievement has sparked discussions about the implications of AI surpassing human abilities in mathematics, with some expressing skepticism about AI's reliability, particularly in generating false proofs or incorrect answers. Concerns are raised about the limitations of AI, especially regarding its ability to handle unsolvable problems or those requiring numerical verification. The conversation also touches on the philosophical aspects of artificial general intelligence (AGI), suggesting that while advancements in AI are significant, true AGI remains a distant goal. Overall, the discussion reflects a mix of admiration for AI's progress and caution regarding its potential pitfalls.
Meow12
Messages
46
Reaction score
20
Technology news on Phys.org
Indeed, but that is not just a press release; it is a peer-reviewed paper in Nature.
Also, in this context I believe DeepMInd is somewhat separate from Google. While they are indeed owned by Google (and their technology is used in Google products) I am not aware of any DeepMind results/paper being retracted. Results published under the DeepMind name seems to have a` good track record.
 
Did they test negatives? What happens when you give it a problem that cannot be proven? Or that can only be verified by numerical evaluation?
 
Swell. A tool that can answer geometry questions incorrectly. Just what we need.
 
  • Like
Likes russ_watters
Vanadium 50 said:
Swell. A tool that can answer geometry questions incorrectly. Just what we need.
It's just that it will certainly come up with a lot of false proofs, it would be nice to have an AI say, "I cannot solve it".
 
Last edited:
From my understanding, the concept of true AGI is still a bit philosophical when we are still scratching the surface of understanding how the human mind works. There will no doubt be continued miraculous developments in AI, but AGI right now is still just a buzzword.
 
Sabine Hossenfelder take on this:
 
  • Like
Likes nsaspook, Meow12 and Greg Bernhardt
Meow12 said:
DeepMind's new geometry problem-solver AlphaGeometry can solve Euclidean plane geometry questions
... AGI just around the corner...

Let's discuss this when there will be an AI capable enough to solve 'neighbour problems' without utilizing shotguns.
Math is easy :wink:
 
  • #10
Rive said:
Let's discuss this when there will be an AI capable enough to solve 'neighbour problems' without utilizing shotguns.
Math is easy :wink:
We've solved that in the UK. The solution is gun laws.
 
Back
Top