#AcademicRunPlaylist - 3/17/25

It's definitely getting warmer, and while trying to adapt to the (relative) heat I was able to go for a short run while listening to talks for my #AcademicRunPlaylist!

First was a fascinating discussion with Andrew Robichaud on America's 19th century ice industry at the Hagley Museum and Library. I previously lived close to Fresh Pond, the center of the global ice trade, so it was even cooler (see what I did there) to learn about this soon-to-be-disrupted industry. Highly recommend https://www.youtube.com/watch?v=BgPLEMyj-EM

Next was an intriguing talk by Zhiyuan Li on incorporating a token reduction method into chain of thought processing at the Simons Institute for the Theory of Computing https://www.youtube.com/watch?v=Ja0V68kIyyM

Next was an excellent talk by Tim Behrens on a cellular basis for mapping behavioral structure at the Kempner Institute at Harvard University https://www.youtube.com/watch?v=JGwyHXigSQc

Next was a thought-provoking talk by Song Mei on a statistical theory of contrastive pre-training and multimodal generative AI at the Simons Insitute https://www.youtube.com/watch?v=wIPkycJg0Vw

Next was an engaging panel on the ethics of variability in hospital policies with Brian Cummings, Lindsay Semler, David Sontag, and Kayte Spector-Bagdady at the Harvard Medical School Center for Bioethics https://www.youtube.com/watch?v=kaTjVrEMQdk

Next was a great talk by Soufiane Hayou on a generative framework for scale-aware training at the Simons Institute https://www.youtube.com/watch?v=S-ChiN395nQ

Next was a nice conversation with Jonas Geiping on using recurrent depth to improve LLM efficiency at TWIML https://www.youtube.com/watch?v=dY90DXLi0vk

Next was an incredible talk by Daniel Beaglehole on detection and steering in LLMs using feature learning at the Simons Institute. By building on earlier feature learning work (which is also amazing) to build concept vectors, Beaglehole shows how these can be used to dramatically steer models. Highly recommend https://www.youtube.com/watch?v=TlLu32yHYfk

Last was a compelling talk by Yue Jiang on computational representations for user interfaces https://www.youtube.com/watch?v=Qulw0ZfdKCo