Is human led mathematics over? Panel with Joelle Pineau, Timothy Gowers & Yann LeCun | Meta AI
Вставка
- Опубліковано 18 жов 2024
- Read about our latest advance in the field of AI and mathematics: bit.ly/3NuYenO
Joelle Pineau, Managing Director of FAIR, sat down for a discussion with Timothy Gowers and Yann LeCun to discuss the current state of field in AI mathematics, the role that it can play in learning and exploration of where they expect the field to go in the future.
--
Subscribe: www.youtube.co...
Learn more about our work: ai.facebook.com
Follow us on Twitter: / metaai
Follow us on Facebook: / metaai
Connect with us on LinkedIn: / metaai
Meta AI focuses on bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.
What a fascinating topic! Thank you so much for sharing this with us ❤️🔥 from Tel Aviv
Thanks for a very enjoyable discussion.
What a fantastic panel.
For answering most of the questions that are raised in this video, I recommend the audience look into random probability theory. This could potentially empower both curiosity+counterintuitivity to AI system.
An absolute treasure. Thank you!
25:40 re breaking down a larger goal into sub goals. I think that's a key point.
Unfortunately I'm not terribly optimistic about Gowers' current approach to ATP. I think the issue is with the "mental models" that mathematicians have, and it's basically completely infeasible to teach these to computers at the moment. These mental models are what makes proving things "obvious" to someone who understands the result very well, while seemingly magic to someone who only sees the symbolic representation.
How can Yann get any work done with that ping pong table in the background?
I disagree with Timothy Gowers in a couple of important points. First of all, proofs that come out of traditional search-based approaches to automated theorem proving are going to be impossible to understand, similarly to how we don't know why Stockfish picks the moves it picks. If anything, using neural networks has the possibility of giving us more human-readable proofs, because we now know how to make them speak natural language.
He seems to be thinking of the situation where an AI is comparable to a human mathematician as some sort of end point. But 5 or 15 years after that the machines will be faster, the techniques will be improved, and an AI will be so much better at math than humans that there will be no competition. Again, look at how quickly computer chess and computer go went from barely at human level to total domination.
The other point is that an AI will be able to solve any IMO problems much earlier than he thinks. There are multiple groups of people actively trying to make LLMs really good at reasoning, and I don't think it's going to take 5 years to see that kind of progress.
2k views for this is actually insulting
I don’t need to watch the video to know the answer to the question in the title is an unequivocal no
also I don’t know if the title is clickbait or stochastic terrorism, but either way it worked
Yes, its over. Mathematics is nothing more than a set of rules which are combined in unique ways to produce new useful results and new rules. These new rules are then added to the old and so on. A computer is suited to just a task and it is only a matter of very short time before AI will easily defeat even the best mathematician. Its going the way of chess.
but not every combination produce useful result
Far from being that. Even logicians don't believe that.