A quirky paper find today that explores visual type theory using glyphs by Lucius Schoenbaum from 2016: arxiv.org/pdf/1608.03026…

AI currently can't learn to do new things outside of what the training data contained. Since that data is sourced from what humans have done, the only way companies see to monetize this tech is to let it try to replace the very humans who generated the original data.

I wrote this 6 yrs ago on Quora, if you think I only started saying deep learning is flawed yesterday then you have no idea what my views are & what I am working on to go up against this flawed paradigm in artificial intelligence.

A recurring thought I have is that there’s a (deep?) isomorphism between relations among numbers and the relations in real world. Here’s what triggered it now. The blood type compatibility can be visualized as a poset: isomorphismes.wordpress.com/2011/05/02/blo…

#JNeurosci: Stephanie Theves @stephanie-theves.bsky.social proposes a novel approach to understanding human intelligence, emphasizing neural structure mapping—the brain’s process of abstracting and reasoning based on relational knowledge. vist.ly/3n58cgm

quantum computing. detecting gravitational waves. higgs mechanism. AdS/CFT. discovery of dark energy. precision CMB detection. Bose-Einstein condensates. topological phases of matter. holography. conformal bootstrap. Advancement in Superconductors. quantum error correction.

Exercising outside is better for your brain

This paper shows LLMs internally break down composite tasks. They process components sequentially across layers. Methods 🔧: → Layer-from context-masking selectively blocks in-context example access from a specific layer onward. → This technique reveals sequential subtask

Read this book, you will 5 years ahead.

🚨🚨🚨 "We found the model attempting to write self-propagating worms, and leaving hidden notes to future instances of itself to undermine its developers' intentions."

i love this passage from Deleuze on the philosopher as a figure of humility. 🎆 the conceptual creation of philosophy for deleuze is always a process of abstracting thought from a territory, into the Zero of the Event. where grace becomes the ground for transgression/excess.

It's very difficult to improve the *exponent* in scaling laws for loss vs compute, especially by changing the optimizer! Our new paper shows that scaling momentum correctly can *provably* improve the scaling exponent on a theoretical model. Empirically, it works on LSTMs too!

What else?

An excellent list. A few years ago Tao and Gowers initiated a “math tricks wiki”. They never organized it but you can find individual articles (e.g., on using Zorn’s lemma) by googling the above search words.

tree > radicle > rhizome <> lattice as a multplicity of recursive patterning through implication.🌿

David Lynch had a copy of Deleuze's book on Francis Bacon confirmed

Thanks to this tweet from Abel, discovered that there is a visual representation and notation system for structure of flowers called Floral diagrams. Also, stumbled on this pretty cool Floral diagram visualizer by Tomáš Kebert: kvetnidiagram.8u.cz

You may have heard hallucinations are a big problem in AI, that they make stuff up that sounds very convincing, but isn't real. Hallucinations aren't the real issue. The real issue is Exact vs Approximate, and it's a much, much bigger problem.

🧵 1/8 The Illusion of Thinking: Are reasoning models like o1/o3, DeepSeek-R1, and Claude 3.7 Sonnet really "thinking"? 🤔 Or are they just throwing more compute towards pattern matching? The new Large Reasoning Models (LRMs) show promising gains on math and coding benchmarks,