Eran Malach
@emalach.bsky.social
Research Fellow @ Kempner Institute, Harvard University
Theory of Deep Learning / Learning of Deep Theory
Theory of Deep Learning / Learning of Deep Theory
Reposted by Eran Malach
In our newest work (led by the amazing
@sunnytqin.bsky.social , w/ @emalach.bsky.social, Samy Jelassi), we investigate a core question for LLMs: "𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘 𝑜𝑟 𝑛𝑜𝑡 𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘" in two prototypical logic-heavy puzzles: CountDown and Sudoku.
@sunnytqin.bsky.social , w/ @emalach.bsky.social, Samy Jelassi), we investigate a core question for LLMs: "𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘 𝑜𝑟 𝑛𝑜𝑡 𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘" in two prototypical logic-heavy puzzles: CountDown and Sudoku.
April 11, 2025 at 4:29 PM
In our newest work (led by the amazing
@sunnytqin.bsky.social , w/ @emalach.bsky.social, Samy Jelassi), we investigate a core question for LLMs: "𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘 𝑜𝑟 𝑛𝑜𝑡 𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘" in two prototypical logic-heavy puzzles: CountDown and Sudoku.
@sunnytqin.bsky.social , w/ @emalach.bsky.social, Samy Jelassi), we investigate a core question for LLMs: "𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘 𝑜𝑟 𝑛𝑜𝑡 𝑡𝑜 𝑏𝑎𝑐𝑘𝑡𝑟𝑎𝑐𝑘" in two prototypical logic-heavy puzzles: CountDown and Sudoku.
Will be presenting this work at #NeurIPS2024, today 11am, poster #2311. Come visit us!
Modern generative models are trained to imitate human experts, but can they actually beat those experts? Our new paper uses imitative chess agents to explore when a model can "transcend" its training distribution and outperform every human it's trained on. arxiv.org/abs/2406.11741
December 12, 2024 at 4:45 PM
Will be presenting this work at #NeurIPS2024, today 11am, poster #2311. Come visit us!
Heading to NeurIPS tomorrow ✈️
Will be presenting a few papers during the week. Ping me if you want to chat!
Will be presenting a few papers during the week. Ping me if you want to chat!
December 9, 2024 at 2:55 PM
Heading to NeurIPS tomorrow ✈️
Will be presenting a few papers during the week. Ping me if you want to chat!
Will be presenting a few papers during the week. Ping me if you want to chat!
Reposted by Eran Malach
I defended my PhD dissertation back in May. I didn't have time to share it widely then (newborn baby), but I think some of you might enjoy it, especially the opening chapters: benjaminedelman.com/assets/disse...
December 2, 2024 at 12:21 AM
I defended my PhD dissertation back in May. I didn't have time to share it widely then (newborn baby), but I think some of you might enjoy it, especially the opening chapters: benjaminedelman.com/assets/disse...
Reposted by Eran Malach
Just put together a starter pack for Deep Learning Theory. Let me know if you'd like to be included or suggest someone to add to the list!
go.bsky.app/2qnppia
go.bsky.app/2qnppia
November 22, 2024 at 9:35 PM
Just put together a starter pack for Deep Learning Theory. Let me know if you'd like to be included or suggest someone to add to the list!
go.bsky.app/2qnppia
go.bsky.app/2qnppia
Reposted by Eran Malach
How does test loss change as we change the training data? And how does this interact with scaling laws?
We propose a methodology to approach these questions by showing that we can predict the performance across datasets and losses with simple shifted power law fits.
We propose a methodology to approach these questions by showing that we can predict the performance across datasets and losses with simple shifted power law fits.
November 21, 2024 at 3:11 PM
How does test loss change as we change the training data? And how does this interact with scaling laws?
We propose a methodology to approach these questions by showing that we can predict the performance across datasets and losses with simple shifted power law fits.
We propose a methodology to approach these questions by showing that we can predict the performance across datasets and losses with simple shifted power law fits.