#AcademicRunPlaylist - 9/19/25

I only had time for a short run, so I pushed my pace to the limit and did a nice circuit while listening to talks for my #AcademicRunPlaylist!

First was a short talk by Anshul Nasery on detecting model misuse through LLM fingerprinting at the Institute for Foundations of Data Science (IFDS). Highly recommend https://www.youtube.com/watch?v=h9j9LfsdXeY

Next was a great talk by Ramya Korlakai Vinayak on learning from diverse human preferences at IFDS https://www.youtube.com/watch?v=W9spOkeZFf4

Next was an excellent talk by Kamalika Chaudhuri on memorization measurement in large models at IFDS https://www.youtube.com/watch?v=otJNJDScuHY

Next was an intriguing talk by Aaditya Ramdas on private evolution's properties at IFDS https://www.youtube.com/watch?v=DA3RsTfDyLw

Next was a compelling talk by Rahul Pahri on global and local minima in shallow neural networks at IFDS https://www.youtube.com/watch?v=OygurZsnk_U

Next was an important talk by Arjun Subramonian on developing a theory of bias amplification in models at the USC Information Sciences Institute https://www.youtube.com/watch?v=R9lFAtm4evg

Next was a thought-provoking talk by Ilias Diakonikolas on learning multi-index models at IFDS https://www.youtube.com/watch?v=qmfUmDrblZ4

Next was a nice talk by Shou-De Lin on neuron-level guidance for steering LLMs at USC ISI https://www.youtube.com/watch?v=uiFS53MRomY

Next was a wide-ranging talk by Samet Oymak on data, architecture, and algorithms in in-context learning at IFDS https://www.youtube.com/watch?v=UuoR37bVy4s

Next was an engaging talk by Nan Jiang on rethinking the theoretical foundation of RL at IFDS https://www.youtube.com/watch?v=6rOxf8NJ4Dw

Last was "Flow" by Mihaly Csikszentmihalyi. When Csikszentmihalyi sticks to his research - the insights of individual task/goal framing and strategy - this book is insightful. Unfortunately, the vast majority of the book is not that, instead philosophizing on the implications of this research for society and the world as a whole. This is complete with blatant racism ("savage" comes out a few times, as does the theory that indigenous Caribbean Islanders died because they were unable to attain a flow state and so lost interest in having sex), as well as less problematic but equally factually incorrect observations (e.g. all animals other than humans are purely instinct driven, money doesn't relate to scientific output, etc.). Some of these, to be fair, were at the time of writing unknown. But Csikszentmihalyi tendency to wildly extrapolate from his research, combined with a complete ignorance of societal and macro forces, fatally handicaps this book. He also commits the cardinal sin of armchair diagnosing public figures, and using anecdotes to indicate that individual framing alone is enough to overcome macro forces, showing literally no quantitative evidence to back up these pronouncements. You're better off reading one of his peer reviewed papers https://www.harpercollins.com/products/flow-mihaly-csikszentmihalyi