By popular demand (n=3), a brief 🧵on anti-concentration. 1. Littlewood-Offord-Erdős ecroot.math.gatech.edu/8803/littlewoo… some applications to random matrices are given here: math.iisc.ac.in/~manju/anti-co…

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

first i thought scaling laws originated in OpenAI (2020) then i thought they came from Baidu (2017) now i am enlightened: Scaling Laws were first explored at Bell Labs (1993)

We're finally out of stealth: percepta.ai We're a research / engineering team working together in industries like health and logistics to ship ML tools that drastically improve productivity. If you're interested in ML and RL work that matters, take a look 😀

there are more things in heaven and earth, Horatio, than are dreamt of in your RL environment

Exciting times!

I like how Timothy Gowers characterized his co-authorship rule:

Daniel Litt byorgey.wordpress.com/2009/01/12/abs…

With Giorgos Vasdekis, we have written a manuscript - arxiv.org/abs/2511.21563 - which surveys the state of affairs within this literature, outlining principles for improving robustness and detailing examples of contemporary methods which confront these issues.

I'll be at NeurIPS this year to talk about self-driving, RL, and all the fun bottlenecks to scaling it up. Come chat with myself or my students: Daphne Cornelisse, Aditya Makkar, Riccardo Savorgnan

1. Ridge regression is heavily used in systematic investing both in the p << n and in the p >> n cases (last one, less so). I don't think that the use is very deeply motivated, other than the old standard argument in favor (see, e.g., El. Stat. Learning).

The poll is now closed - the correct answer is: nearly 100% The players almost always end on the same digit! just under 1/3 of you got it correct. The reason is coupling (see included details), giving about a 97.5% chance of them ending on the same digit. Also: see simulation

New YouTube video “The Joy of Coupling” Happy Valentine’s Day! (link below)

1/ Technical thread on #1stProof Problem 6: finding “spectrally light” vertex subsets in a graph, and how its solution fits into the landscape of spectral sparsification + restricted invertibility. Original thread: x.com/yangpliu/statu…

Wasserstein geometry = quotient geometry of permutation invariance. In this blog, I explain why this is the natural language for exchangeable particles—and why mean-field neural network training shows up as a W2 gradient flow. mufan-li.github.io/OT2/

At this point LLMs are smarter than the people who call them stochastic parrots.