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: 🧵

Möbius transforms and Shapley values for vector-valued functions on weighted directed acyclic multigraphs

We generalize the concept of Möbius inversion and Shapley values to directed acyclic multigraphs and weighted versions thereof. We further allow value functions (games) and thus their Möbius transform...

arxiv.org

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.

Abel Jansma@abelaer.bsky.social · 11mo ago

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 #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):

Abel Jansma@abelaer.bsky.social · 2y ago

🚨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 🧵

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!

Abel Jansma@abelaer.bsky.social · 2y ago

🚨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 🧵

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...

Abel Jansma@abelaer.bsky.social · last yr.

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

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

A Mereological Approach to Higher-Order Structure in Complex Systems: from Macro to Micro with Möbius

Relating microscopic interactions to macroscopic observables is a central challenge in the study of complex systems. Addressing this question requires understanding both pairwise and higher-order inte...

arxiv.org

Abel Jansma@abelaer.bsky.social · 2y ago

🚨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 🧵

🚨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!

Chris Ponting@cgatist.bsky.social · 2y ago

📢Now published! Know that clustering cells in reduced dimensions is problematic? Want to identify cell subtypes or states *without* clustering? Then check out Stator which finely resolves states & subtypes even among apparently homogeneous cells Thread www.embopress.org/doi/full/10....

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!

EASTBIO: Mapping DNA variants to cell states in health and disease at University of Aberdeen on FindAPhD.com

PhD Project - EASTBIO: Mapping DNA variants to cell states in health and disease at University of Aberdeen , listed on FindAPhD.com

shorturl.at

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?

Veera Rajagopal@doctorveera.bsky.social · 2y ago

Association of polygenic scores for neuropsychiatric traits with self-reported professions based on analysis of 420k individuals from UK Biobank and Million Veteran Program. Look at the 'arts & design' category. Artistic talent comes at a cost--a piece of your mind :)

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

Friderike Spang@friderikespang.bsky.social · 2y ago

"Washes up to 2x more than expected"—philosophers, I need your counsel on this serious matter! Is this a self-defeating promise? By promising to exceed expectations, does this detergent raise those expectations to a level it cannot beat due to its very promise of beating them? 🤔 #PhilSky

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...