What problem is explainability/interpretability research trying to solve in ML, and do you have a favorite paper articulating what that problem is?

I’ve been preparing a presentation on this paper and trying to more concisely get the main point across. I’d say it comes down to this: To make the interpretive lens of a particular modeling decision visible, you have to contrast it against a population of alternative decisions. 1/5

if attention mechanisms in transformers create graphlike transformations across text then it is actually an incredible pun that groups of sentences are called "para-graphs"

"Helen de Cruz (1978–2025): an academic and personal tribute" *Religious Studies* Published online 2025:1-4. doi.org/10.1017/S003... @helendecruz.net

Extremely good stuff from Leif here

I really enjoyed talking to @nsaphra.bsky.social about her thoughts on what much language model interpretability research misses. My latest in @quantamagazine.bsky.social:

i think regardless of who has authored a paper, we (the scientific community) can and should stop criticizing the work on lack of perceived novelty. we can do better than perpetuating harmful myths about individual scientific papers having to present a discovery no one has ever thought of before.

Forgive one last point to vent about: The reviewer’s only other criticism was a lack of novelty. The obsession with novelty has made a lot of ML work insufferable so it’s unfortunate to see it here. But here’s who is apparently already making our points about how ML is used for humanities research:

How to quantify the impact of AI on long-run cultural evolution? Published today, I give it a go! 400+ years of strategic dynamics in the game of Go (Baduk/Weiqi), from feudalism to AlphaGo!

Well, this got rejected today as a short position paper from CHR so I suppose I can share it now: arxiv.org/pdf/2508.00095 Here’s a hopefully coherent ramble about what we’re trying to make clear:

Well, this got rejected today as a short position paper from CHR so I suppose I can share it now: arxiv.org/pdf/2508.00095 Here’s a hopefully coherent ramble about what we’re trying to make clear:

Out now! Bride of Frankenstein [film|minutes]

A quick summary for busy people: We propose a mathematical framework for learning representations by extracting the algebra of transformations of worlds from the agent's perspective. 1/n www.sciencedirect.com/science/arti...

💻 We are very excited to announce the launch of our special issue Computational Formalism, edited by Tess McNulty and Laura Alice Chapot, that reconsiders the intersections of computation and form at this emerging technological and critical moment. culturalanalytics.org/issue/12788

“[Something] unpredictable, spontaneous, unformulable and ineffable is found in any terminal object. Standardizations, formulae, generalizations, principles, universals, have their place, but the place is that of being instrumental to better approximation to what is unique and unrepeatable.” (1/2)

A GoFundMe for the instructor who was fired at Texas A&M!

I have a thing under review right now that argues translations across research domains (as epistemic “languages”) are useful insofar as translators can give a theory of their work. Auto-translation necessarily hides interpretive choices and so can only be useful in very narrow cases.

Why do societies reliably develop strikingly similar traditions like dance songs, hero stories, shamanism & justice institutions? In a new BBS target article, I propose a theory for such "super-attractors" + cultural evolution more broadly. Now open for commentary: www.cambridge.org/core/journal...

Great bit from Leslie Lamport in this video: “Coding is to programming what typing is to writing. Writing is something that involves mental effort. You’re thinking about what you’re going to say. The words have some importance but in some sense even they are secondary… youtu.be/rkZzg7Vowao?...

Here's a nice new paper yet again showing the benefits of drawing for better comprehension and understanding (which is not surprising from a multimodal language view): "Benefits from sketching for improving comprehension monitoring from illustrated texts" link.springer.com/article/10.3...

Well, there’s the answer to my student’s question

You will never be able to make "data driven" decisions until you realize just how much your decisions are driven by emotions: fear of missing out, defensiveness of your favorite idea, apprehension over something unusual. That's because there's no such thing as "the data" - only interpretations.

Seeing that @matthematics.com liked a post is a nice way to find out he’s here on Bluesky. I’ve been watching his abstract algebra videos over the summer and they are unbelievably good. Highly recommend.

Left out a crucial step (to not do): (2a) Evaluate the process by which the numbers in (2) were computed according to a metric that has no relevance to (1).

How to not do computational humanities: (1) Lay out a question/hypothesis about a complex, cultural domain. (2) Compute numbers that were inspired by (1) but without sufficiently formalizing (1) so as to meaningfully link to it. (3) Interpret numbers to mean whatever your prior vibes were.

Chess 2 *finally* got released.

plant_pol_kato: Kato plant-pollinator web. A bipartite network of plants and pollinators from Kyoto University Forest of Ashu, Japan, from 1984 to 1987. 772 nodes, 1206 edges. https://networks.skewed.de/net/plant_pol_kato Ridiculogram:

At long last, I can post my team's summer project: applied modules to teach how ML/AI tools are changing social science and humanities research: ubcecon.github.io/praxis-ubc/ Highlights: LegalBERT to analyze anti/pro-immigrant sentiment in 19th c. BC law: ubcecon.github.io/praxis-ubc/d... 🧵1

I get the sense that a fair number of people attribute a kind of strong emergence to LLMs such that there’s a hard distinction between an LLM as giant function and the simple functions that constitute it.

Is anarchist science possible? As an experiment, we got together a large group of computational neuroscientists from around the world to work on a single project without top down direction. Read on to find out what happened. 🤖🧠🧪

Hiroki Sayama announced that CCS2026 will be held in Binghamton, USA — see you there! 🇺🇸 #CCS2026

Jim and I previously collaborated on a project in which we tried out some ways of talking about the differences between religion-focused subreddits through the differences in topic models trained on their text. Since then we revisited our basic assumptions on what this kind of work can tell us. 1/

Shit.

I'm thrilled to be doing an oral presentation on "Culture is not Trivia" at #ACL2025 next Wednesday 7/30, as well as participating in the human-centered NLP panel afterwards! (thanks also @lauraknelson.bsky.social for the shoutout in her #ic2s2 keynote today!) aclanthology.org/2025.acl-lon...

I can't* fathom why the top picture, and not the bottom picture, is the standard diagram for an autoencoder. The whole idea of an autoencoder is that you complete a round trip and seek cycle consistency—why lay out the network linearly?

What makes a puzzle hard? I found this video delightful.

I spent the past year slowly working through Eduardo Kohn’s book, How Forests Think: Toward an Anthropology beyond the Human. By using a non-human medium to study humans, I really think it has a lot to offer to those of us trying to use math/computation to understand humans.

New paper out titled, The differential meaning of models: a framework for analyzing the structural consequences of semantic modeling decisions. Co-authored with Jim Deitrick. We try to say some things about semantic models in simple, precise terms. Read it here: arxiv.org/abs/2509.00248

Jim and I previously collaborated on a project in which we tried out some ways of talking about the differences between religion-focused subreddits through the differences in topic models trained on their text. Since then we revisited our basic assumptions on what this kind of work can tell us. 1/

New paper out titled, The differential meaning of models: a framework for analyzing the structural consequences of semantic modeling decisions. Co-authored with Jim Deitrick. We try to say some things about semantic models in simple, precise terms. Read it here: arxiv.org/abs/2509.00248

My drawn statement on Ai as standalone from my now finished minicomic as syllabus for new liberal studies class! As promised this is shareable, printable - all from my site, feel free to make use of it, cite me & let me know how it’s received. Share away all here! spinweaveandcut.com/fall-2025-sy...

I have a new book coming out this autumn via @mitpress.bsky.social ...details at: mitpress.mit.edu/978026255348... #HistSci #Computing #History #Books 🗃️

What is emergence? This is one of the most central ingredients of complexity, and a challenging one to formalize. Here's a paper by @seanmcarroll.bsky.social & Achyth Parola that attempts to classify different forms of emergence. @manlius.bsky.social @sfiscience.bsky.social arxiv.org/pdf/2410.15468

I am thrilled that my paper on machine learning in science and the myth of theory-free inference has been accepted for publication in Erkenntnis. The paper remains available for download as a preprint through the archives. philsci-archive.pitt.edu/26075/1/Andr...

I’m excited to announce that my new book, _The Irrational Decision_, is now available for pre-order from Princeton University Press.

Korzybski's maxim, 'the map is not the territory' is conventionally thought of as an epistemological warning about confusing symbols with reality. But what if the organism is always already a map? This transforms the maxim into a fundamental statement of ontology. 🧵 1/7

A little empirical paper on LLMs I worked on is now up on PNAS, studying LLMs' tendency to prefer goods and services pitched in LLM prose: www.pnas.org/doi/10.1073/...