First Author Interview: AI & formal math (Formal Mathematics Statement Curriculum Learning)

First Author Interview: AI & formal math (Formal Mathematics Statement Curriculum Learning)

Mar 08, 2022
|
24 views
Details
#openai #math #imo This is an interview with Stanislas Polu, research engineer at OpenAI and first author of the paper "Formal Mathematics Statement Curriculum Learning". Watch the paper review here: https://youtu.be/lvYVuOmUVs8 OUTLINE: 0:00 - Intro 2:00 - How do you explain the big public reaction? 4:00 - What's the history behind the paper? 6:15 - How does algorithmic formal math work? 13:10 - How does expert iteration replace self-play? 22:30 - How is the language model trained and used? 30:50 - Why is every model fine-tuned on the initial state? 33:05 - What if we want to prove something we don't know already? 40:35 - How can machines and humans work together? 43:40 - Aren't most produced statements useless? 46:20 - A deeper look at the experimental results 50:10 - What were the high and low points during the research? 54:25 - Where do we go from here? Paper: https://arxiv.org/abs/2202.01344 miniF2F benchmark: https://github.com/openai/miniF2F Follow Stan here: https://twitter.com/spolu Abstract: We explore the use of expert iteration in the context of language modeling applied to formal mathematics. We show that at same compute budget, expert iteration, by which we mean proof search interleaved with learning, dramatically outperforms proof search only. We also observe that when applied to a collection of formal statements of sufficiently varied difficulty, expert iteration is capable of finding and solving a curriculum of increasingly difficult problems, without the need for associated ground-truth proofs. Finally, by applying this expert iteration to a manually curated set of problem statements, we achieve state-of-the-art on the miniF2F benchmark, automatically solving multiple challenging problems drawn from high school olympiads. Authors: Stanislas Polu, Jesse Michael Han, Kunhao Zheng, Mantas Baksys, Igor Babuschkin, Ilya Sutskever Links: TabNine Code Completion (Referral): http://bit.ly/tabnine-yannick YouTube: https://www.youtube.com/c/yannickilcher Twitter: https://twitter.com/ykilcher Discord: https://discord.gg/4H8xxDF BitChute: https://www.bitchute.com/channel/yannic-kilcher LinkedIn: https://www.linkedin.com/in/ykilcher BiliBili: https://space.bilibili.com/2017636191 If you want to support me, the best thing to do is to share out the content :) If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this): SubscribeStar: https://www.subscribestar.com/yannickilcher Patreon: https://www.patreon.com/yannickilcher Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2 Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n

0:00 - Intro 2:00 - How do you explain the big public reaction? 4:00 - What's the history behind the paper? 6:15 - How does algorithmic formal math work? 13:10 - How does expert iteration replace self-play? 22:30 - How is the language model trained and used? 30:50 - Why is every model fine-tuned on the initial state? 33:05 - What if we want to prove something we don't know already? 40:35 - How can machines and humans work together? 43:40 - Aren't most produced statements useless? 46:20 - A deeper look at the experimental results 50:10 - What were the high and low points during the research? 54:25 - Where do we go from here?
Comments
loading...