I did a spot check of my most recent math question I asked on math forums against LLMs Gemini Flash Thinking: 2/2 DeepSeek: 2/2 Claude: 1/2 ChatGPT o1: 0/2 Gemini was the fastest, followed by DeepSeek, followed by human. docs.google.com/document/d/16S…

Sitting in airport and realizing I don't get how singular values work. Does anyone understand why the two graphs match? mathoverflow.net/questions/4848…

Fun visualization: take 10x10 random matrix and visualize eigenvalue trajectories you get by rotating this matrix by theta in 0..2 Pi in some fixed direction. Now vary this direction smoothly

From a talk by Chris Manning

Are there recorded talks I can watch relevant to DeepSeek?

Keeping up with headline news, which are often negative, makes it easy to lose track of the big picture

I'm giving a talk at the MCDC🤝 workshop (#ICLR2025) tomorrow! Planning to cover: * An overview of decentralized DL & its links to other fields * Lessons learned from research on Learning@home, DeDLOC, SWARM, Petals * Sneak peek on some of our upcoming work! See you at 14:30!

Watching Zhuang Liu's - "Transformers without Normalization", this slide is a reminder how our optimizer and architecture choices are coupled

Unexpected RMT observation, squared singular values of a product of random projections are essentially distributed as exponentiated chi-squared, can anyone see a direct explanation of this? math.stackexchange.com/questions/5060…

Once everyone online is indistinguishable from an AI agent, it would make it cool again to hang out in person. Until the robot impersonators.

Enjoyed Jeremy Bernstein thought-provoking talk on optimizers at ML Collective today. Are theories that motivate optimizers very useful? Adversarial for AdaGrad, natural gradient for KFAC. Non-linear solvers in scientific computing seem to advance without spending a lot of effort thinking

came across this overview by Derek Lowe on the state of AI drugs a year ago ACS Central Science science.org/content/blog-p…

Was brushing up on transforms for some scaling laws math, summarized here community.wolfram.com/groups/-/m/t/3…

This is a nice explanation on why reasoning emerges as an unexpected side effect of training for text compression (but not video compression)