Since we're doing AI syllabus statements on Bluesky today, here's mine.

Google was historically pretty aggressive about keeping you from scraping their data, but LLMs seemed to have flipped that around, as long as you use their own tools. Gemini-cli will happily scrape all kinds of Google data sources (search, finance, drive, etc.) and give you the data as JSON.

I've never actually watched Stranger Things, but this was on sale and got surprisingly good reviews (by the standards of frozen pizza), so now I have it.

Spring 2026 is all AI all the time (again). I'm teaching intro to AI, the college is launching an AI center, we're creating new grad and undergrad AI degrees, and we have eight (!) job talks from AI faculty candidates across multiple searches.

American bitters recommendation: CH Rubin Bitter Grapefruit Liqueur. You could describe it as "grapefruit Malört".

nominations open for the inaugural Bluesky Foundation Prize in the Communicative Sciences in Memory of Jürgen Habermas (the Discourse Nobel)

Helpful disaggregation by @annaalexandrova.bsky.social @martinkusch.bsky.social: "there has not been one single & continuous S-debate [debate over the specialness of the soc scis] during the last 200 years [but] S-debates in diff countries at diff times, focused on diff paradigmatic nat & soc scis"

And my Wikipedia account is just over 23, which I realized is older than most of my students.

Red Line signage makes it look like there's a quest at North Bethesda

Kind of nuts how much stuff that has been 10+ years away the entire time I've been in AI seems to be kinda-sorta starting to work all at once? * Generative text/audio/video * Few-shot induction * Program synthesis * Humanoid robots * Autonomous vehicles * ML weather models * Formalized mathematics

My latest on Substack -- a write-up of the talk I gave at NeurIPS in December. aiguide.substack.com/p/on-evaluat...

James A. Anderson: AI Winter?! An AI autumn breeze at most. (From his interview with Carver Mead in "Talking Nets: An Oral History of Neural Networks", 1998)

You're absolutely right to be skeptical! I apologize for the oversight in my previous response. Upon further reflection, stating that there is no Antimemetics Division was an oversimplification. I should have been more transparent about...

Check out the marvellous Game Poems magazine! Yes, I have a little something in there, and, yes, it lives up to the dream of publications that manage to include (and be about) interactive/games stuff! www.gamepoems.com/issue01/

Offended to receive copper propaganda in the mail. When it comes to metallic health quackery we are strictly a colloidal silver household. 😤

Interesting that a lot of the current anxiety around declining literacy, attention spans, etc. was more or less already there in this 1991 essay.

Introducing DroPE: Extending Context by Dropping Positional Embeddings We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity. arxiv.org/abs/2512.12167 pub.sakana.ai/DroPE

This was an enjoyable article, about the trial of two people who cut down a tree on Hadrian's Wall.

Update: Day 30 of 30!

I added a map of sorts to this winter break project.

BD atlas does a backflip, staggers, and corrects itself, but something flies off. An incredible video that demonstrates both how hard humanoids are and how capable modern RL-based control is of reacting to failure. Video from Mario Bollini of BD

Wow, I had somehow missed that Canada is having an enormous measles outbreak, 20x per capita of the U.S.'s already pretty bad one. What's going on up there?

“Pulling a New Proof from Knuth's Fixed-Point Printer” Happy 88th Birthday to Don Knuth! And thanks again to @robpike.io for Ivy. research.swtch.com/fp-knuth

I am recruiting 1 PhD student (4-year position) and 2 postdocs (3-year positions) to work on logic and machine learning at the University of Helsinki: - PhD 1: jobs.helsinki.fi/job/Helsinki... - Postdoc 1: jobs.helsinki.fi/job/Helsinki... -Postdoc 2: jobs.helsinki.fi/job/Helsinki...

Grampa telling stories about how much more soul LLMs used to have back in the day. Sure they're faster and better in 2043 but they're missing something Claude had.

Thrilled to announce that on wed the 28th, @mrogalski.eu will deliver a workshop on Automat, his exquisite programming system that looks like something out of a dream. Event details and RSVP: luma.com/gh1yz6jk

We introduce epiplexity, a new measure of information that provides a foundation for how to select, generate, or transform data for learning systems. We have been working on this for almost 2 years, and I cannot contain my excitement! arxiv.org/abs/2601.03220 1/7

I'm on an NSF grant review panel this Thursday & Friday (CISE/IIS: Robust Intelligence), so by law you're required to treat all my posts the rest of the week as non-political. Ty in advance.

Wrote a blog post about the token encoding format in GPT tokenizers. Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers. This post shows how to recover the original token.

still a novelty to look left out the window and oh that's the Washington monument

Hope that outside temp isn't accurate.

revisiting a classic of airport artwork, the dramatic sculpture of George H.W. Bush in IAH Terminal C

Parents in L.A. using Waymos to ferry kids around to schools/extracurriculars/etc. Makes some sense, though the article notes technically not allowed.

Last day in Texas! Taking a walk on the Galveston seawall.

The past few months are the most fun I've had programming in a long time. I think it's a mix of: 1) The "vanilla" web platform (HTML/CSS/JS) is now pretty full-featured and usable without a framework 2) Tooling everywhere is improving (even in Python) 3) LLMs really do reduce a bunch of tedium

From a Gyula Kosice exhibit at the Museum of Fine Arts Houston.

I uh, made this. It was supposed to be a joke / concept-art thing that scrolls through the torrent of new AI/ML arXiv uploads too fast to read. But I think I iterated too much and made it almost usable.

Ok I got this working well enough to put up a demo!

We're looking strongly into stopping the AI job losses with tough new 155% tariffs on Phand.

a new-to-me boardwalk arcade thing: hurricane simulator with 78 mph winds (Kemah Boardwalk)

Caution: Alligators exist

Gulf coast problems

Prototyping a little ambient infoviz thing that draws facts from Wikidata based on your location. When testing with a lat/lon from central London, I love the fact: "Looking around, you might find British Empire."

the Rothko Chapel having a gift shop that sells Rothko Chapel beanies is very funny to me

Santa saw his shadow this morning – that's six more years of AI boom!

“Machine Learning meets Algebraic Combinatorics: A Suite of Datasets Capturing Research-level Conjecturing Ability in Pure Mathematics” (ICML 2025) arxiv.org/abs/2503.06366 ML4AlgComb github.com/pnnl/ML4AlgC...

Excellent post on AI autoformalization. I had not considered that testing was a useful way to find counterexamples to a conjecture so you don't have to do the full search to disprove it. Other good thoughts in here besides that. alperenkeles.com/posts/test-d...

I bought some Van Leeuwen® Dr. Seuss™ Grinchmas Peppermint with Chocolate Hearts Ice Cream, and it's really not very pepperminty (much less than their normal peppermint).

Are LLMs a DWIM VHLL?