#AcademicRunPlaylist - 7/12/24

It felt like running in a sauna, but I was still able to go on a decent run while listening to talks for my #AcademicRunPlaylist!

First was a fascinating talk by Michael Levin on the surprising range of biological intelligences at UQAM | Université du Québec à Montréal https://www.youtube.com/watch?v=2MlIHqWITmU&t=5s

Next was a great panel on private law and AI and the UCL Faculty of Laws with Ernest Lim and Phillip Morgan. When asked at the end of the panel whether they used generative AI to write their new book, Morgan's response is instructive: "It's the best of my knowledge that [generative AI] was not used in the editing or writing of any of the chapters, nor in my view should it be." 🔥 https://www.youtube.com/watch?v=rS0lwwrh4qI

Next was an interesting talk by Christian Lebiere on integrating LLMs into cognitive architectures at UQAM https://www.youtube.com/watch?v=68uBFQjzPrk

Next was an excellent talk by Salvatore Lattanzio on how factory automation shapes the allocation of workers across firms at the Faculty of Economics, University of Cambridge https://www.youtube.com/watch?v=g4FdmXa-P28

Next was an amazing talk by Jackie Chi Kit Cheung on benchmarking and evaluation in NLP at UQAM. Cheung gives a thorough examination of different approaches, highlighting their strengths and weaknesses and the particular difficulty of benchmarking/evaluating generative systems. Finally he details a rigorous approach for benchmark creators to specify how they should be used, which should help everyone cut through the huge noise in the space. Highly recommend https://www.youtube.com/watch?v=h1-vMTFtl1c

Next was an engaging panel on the current state of the US workforce and labor economics at the New York University School of Law with Dan Hamermesh and Ye Zhang. Hamermesh had a particularly incisive zinger: "My fellow alum from Chicago, Bernie Sanders, proposed a mandatory four day, 32 hour week. Of course, this is not gonna happen, like anything else Bernie has proposed." https://www.youtube.com/watch?v=dingpUUPqTA

Next was a fantastic talk by Alexei Efros on the primacy of data in driving recent advances in AI at UQAM. Efros gives one of the best accounts for why data deserves a huge amount of credit for driving improvements in AI, clearly demonstrating how even simple methods, when given enough data, match and sometimes outperform more elaborate models. Quote of the talk: "Interpolation in sufficiently high dimensional space might be indistinguishable from magic." Highly recommend https://www.youtube.com/watch?v=R7qy2BY6mTk

Next was an important panel on religious bias claims in the workplace given the recent Supreme Court decision in Groff v. DeJoy at the NYU School of Law with Marjorie Mesidor, David Sherwyn, and Zoe Salzman. It's a pretty depressing but vital conversation, and listening to these legal experts try to make sense of this decision at least made me feel like I wasn't crazy for also not understanding the underlying logic https://www.youtube.com/watch?v=A0Pj6cJWZqE

Last was an intriguing talk by Sonia Chernova on using LLMs to help robots navigate the diversity of human home environments at Princeton Computer Science https://www.youtube.com/watch?v=i3z_7hNiOUc