New paper alert 🚨 “A Compositional Calculus for Semantic Synergy in Language Model Embeddings” A training-free way to extract synergy from LLM embedding vectors. 🧵

Gave a talk at the Softmax Alignment meeting—pretty nice lecture room.

I asked somebody in the Berkeley AI safety scene if they meditate. “No,” they said, “I’m not really bliss bottlenecked at the moment”.

Every year, our apricot is the first to blossom, which is easy to remember because “apricot” shares a linguistic root with “precocious”.

I hope that when I'm 88 I'm also still able to embrace new technologies with such joy and curiosity: www-cs-faculty.stanford.edu/~knuth/paper...

I found this surprising and somewhat worrying: openAI is using an AI model to align their model spec. #NeurIPS2025

A few months ago I started discussing causal emergence with Erik Hoel. This led to a really fun collaboration, and a new approach to “engineer emergence”. Erik just published an overview of the ideas, goals, and dreams:

Many sea stars begin life as young fairy-like creature (called a brachiolaria) that float through the open ocean. Eventually, a small star forms within them (here in yellow). The fairy-like brachiolaria sinks under the star’s weight, and the star pops out! 🎥@the_story_of_a_biologist (on Insta)

🚨New paper: arxiv.org/abs/2510.05786 *Shapley values beyond game theory* We show that Shapley values aren’t just about dividing payoffs--they are the right way to project down any higher-order structure. We generalise them, and Möbius inversions, in important ways: 🧵

This is how Rota originally introduced the incidence algebra. Everyone since has (correctly) required the ring to be commutative. Did people in the '60s just refer to commutative rings as associative rings?

Just to balance out the discourse: the #NeurIPS2025 review process for this paper went great. Fair reviews, mostly responsive reviewers, and a thoughtful AC that caught a possible conflict of interest. Definitely improved the paper.

The #NeurIPS2025 version is now online: arxiv.org/pdf/2501.11447 It includes a new analysis to show that LLM semantics can be decomposed: the negativity of "horribly bad" is redundantly encoded in the two words, whereas "not bad" has synergistic semantics (i.e. negation):

The "partial causality decomposition" was just accepted for a spotlight at #NeurIPS2025! The final version includes a decomposition of LLM semantics---the Arxiv version should be updated soon. Stay tuned!

While I'm flattered, it's a bit weird that google's AI defers to me when you search for this:

Next week we're organising a workshop on the role of analogies in (artificial) intelligence, with: Melanie Mitchell (@melaniemitchell.bsky.social), Martha Lewis, Jules Hedges (‪@julesh.mathstodon.xyz.ap.brid.gy‬), and Han van der Maas. Register here: www.d-iep.org/workshopanal...

Don't take my word for it--take Reviewer 2's: "I found the paper extremely interesting and deep" A gentle introduction is available at abeljansma.nl/2025/01/28/m...

My new approach to higher-order interactions in complex systems is now published in Physical Review Research: journals.aps.org/prresearch/a...

just one more index bro I swear just one more subscript and it's gonna be so clear just one more index please bro

New blog post 🚨 A gentle dive into the mereology of complex systems, Möbius inversion, and a new way to think about higher-order interactions—no prior knowledge needed! abeljansma.nl/2025/01/28/m...

New year, new Möbius inversion! This new paper is an example of my more general proposal that you should study complex systems with 'quantitative mereology' by applying the Möbius inversion theorem: arxiv.org/abs/2404.14423

🚨New paper! The “partial causality decomposition” When many things influence an outcome, how to distinguish individual vs collective #causality? A framework to disentangle these complex relationships into synergistic, redundant, and unique components: arxiv.org/abs/2501.11447 🧵

Very excited that this is now published: Stator is a new method to infer higher-order interactions in gene expression data, which reveals cell states and types without clustering! It's an "intersectional" approach to cellular identity: cells can have multiple identities--states are not disjoint!

this is a pretty interesting way of looking at collatz math.stackexchange.com/questions/26...

I bet those ants can get an even bigger sofa around a corner.

There’s an analogy to be made between Wolfram’s physics project, and Borges’ Library of Babel.

In between changing countries, jobs, and becoming a father, I didn’t manage to read much this last year. But I just finished “the Maniac” by Labatut and thought it was fantastic. On to “the world beyond the world” by @erikhoel.blogsky.venki.dev.

I am STOKED that myself and @avakhamseh.bsky.social are looking for a joint EastBIO PhD candidate across @uniofaberdeen.bsky.social and @uoe-sps.bsky.social to develop cutting-edge statistical models at the interface of single cell genomics and quantitative genetics: shorturl.at/vQo0P pls reskeet!

Is it a common assumption that personality traits like openness and agreeableness are genetically determined? Or is this circular reasoning because their polygenic scores are derived from the UKBB in the first place?

Hofstadter's laundry: this soap washes 2x more than expected, even when you take into account Hofstadter's laundry

Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @msftresearch.bsky.social ch AI for Science. www.biorxiv.org/content/10.1...

omg I just found out that cod(f) ≠ dom(g) tfw no gf😔

I always found it crazy that there were just 66 years between the first human flight and the moon landing. Now in 15 months, AIs have gone from random guessing to expert level.

Swift, 1887

A lot of emergence research goes: 1) Here's a new measure for emergence. 2) It makes sense on 3 deterministic var's. 3) We gave 6 people ketamine and ayahuasca and asked them about their parents while hooked up to an EEG machine. 4) Emergent brain dynamics! I feel like there's a step 2.5 missing.

Is this @ricardsole.bsky.social drawing a hopfield swarm of tumbleweeds?

I’ve had the same experience. Is this mainly a difference in the human feedback part of RLHF? If so, then we should create a version where it’s just scientists giving feedback for this specific use case: thinking through and formalising ideas.

I’m not affiliated to this group but have only seen and heard great things from them, so this looks like a very cool position!

How your email finds me

Write math on 🦋 with UnicodeIt! For example: θ ∈ ℝⁿ or pp̅ → μ⁺μ⁻ Use website or install system-wide in Linux, macOS, or windows www.unicodeit.net (Created several years ago with @svenkreiss.bsky.social)

I couldn’t find a starter pack for emergence researchers, so i started one. Which bsky accounts am I missing? go.bsky.app/G8Abzar

I guess this is true for ML in general, since the whole field essentially evolves by gradient descent, so it doesn’t matter if some unknown number of claims are false. However, as an individual researcher, I still want to know if *my* claim is statistically valid.

This is quite surprising. Is this the result of the starter pack feature?

Ice skating diatoms!

I seem to be following…myself? Is this supposed to be possible? @bsky.app

Looking for the starter pack that contains all starter packs

When I come across papers like this, my first thought is “so cool/quaint!”—quickly followed by “yeah I’m not gonna read this”. Latex has made me typographically xenophobic.

New preprint! “The Fast Möbius Transform: An algebraic approach to information decomposition” Deep theory of higher-order information theory with immediate practical applications and speedups—a very fun collab with @frosas.bsky.social and Pedro Mediano arxiv.org/abs/2410.06224 1/n

🖐️ New here—I’m a physicist, trying to understand the role of composition, hierarchies, and higher order interactions in complex systems. I hope to create bridges between #physics, #biology, #machinelearning, #AI, #cybernetics using #mathematics: arxiv.org/abs/2404.14423 Stay tuned for more on this!