#AcademicRunPlaylist - 11/19/24

I was able to escape for a bit and go out for a nice run accompanied by talks for my #AcademicRunPlaylist!

First was a slate of talks from the Domain Adaptation workshop at the Simons Institute:

Yuekai Sun (transfer learning for weak-to-strong generalization) - https://www.youtube.com/watch?v=q68VtFoWmFI

Yi Yu (differential privacy in distributed learning) - https://www.youtube.com/watch?v=tboTpB4RVcI

Mehryar Mohri (discrepancy-based theory of adaptation) - https://www.youtube.com/watch?v=QOQYid8P3f8

Sara Beery (distributional shift in ecological data) - Beery dissects general problems with distribution shift research and the illusion of recent progress in the space, highly recommend - https://www.youtube.com/watch?v=8EfQ4C--WD0

Victor Veich (off-target behavior in LLM alignment) - Veitch provides a nice formalism for alignment, showing how one can unify different aspects of alignment to train more desirable models, highly recommend -  https://www.youtube.com/watch?v=kPhlpqxAN7c

Maggie Makar (causally motivated robustness to shortcut learning) - https://www.youtube.com/watch?v=oj7wXyKKbYQ

Tengyu Ma (extrapolation of models to unseen domains) - https://www.youtube.com/watch?v=7raKN7zofno

Jacob Steinhardt (scalably understanding LLMs) - Steinhardt presents an impressive method to find prompt inputs that give desired model outputs, demonstrating that this method can transfer from open to closed models. Highly recommend https://www.youtube.com/watch?v=KjOankEWYgk

Next was a great talk by Jaime Fernández Fisac on building safe autonomous driving systems at the MIT Embodied Intelligence seminar. Includes the gem that current autonomous vehicles would need to be on the road for 500 years to collect enough data to provide safety guarantees equivalent to human drivers  😬 https://www.youtube.com/watch?v=c3MCCUdbKPw

Next was an excellent conversation with Thomas Höppner on the need for more robust competition law enforcement, the nascent effects of the DMA, and more on the Digital Markets Research Hub https://www.youtube.com/watch?v=Jmr8NLREBv0

Next was an interesting talk by Jason Ma on using large models for robot training supervision at MIT https://www.youtube.com/watch?v=JfZYtpEisoM

Next was a fantastic talk by Michael Tomasello on the evolution of agency at the Kempner Institute at Harvard University. Tomasello convincingly demonstrates the different types of agency likely exhibited by our ancestors, how people differ, and the implications for human development. Highly recommend https://www.youtube.com/watch?v=d_kzQvVI8rk

Next was an amazing talk by Supreet Kaur on the role of habit formation in labor supply at CEPR. Using a unique experiment in a Chennai labor market that incentivized subjects to show up daily, Kaur shows a persistent effect after incentives are removed (unless there is an individual shock). Highly recommend https://www.youtube.com/watch?v=s_JKzZjxOXE

Next was an engaging talk by Virginia Smith on the current state of AI safety at UC Berkeley https://www.youtube.com/watch?v=p35cSCMpyBY

Next was a nice talk by Lionel Robert on worker perceptions of robots in the restaurant industry at Stanford https://www.youtube.com/watch?v=qNKh_pKKgC8

Next was a short talk by Bill Marino on automated EU AI Act compliance analysis at Stanford CodeX https://www.youtube.com/watch?v=GwBjfp_O4HY

Last was a compelling talk by Rediet Abebe on algorithmic prediction in education and frameworks for assessing when individual-level prediction is truly necessary at the Schwartz Reisman Institute https://www.youtube.com/watch?v=FPbg9_dKARk