Collaborative groups often outperform single individuals in complex problem solving. A new paper examined how to create the right incentives to promote this kind of collective intelligence. www.pnas.org/doi/epdf/10....

I am flabbergasted I am by how much vibe coding has expanded my capacities as a scientist and teacher. In the last few weeks, I've mocked up class demos of a live turing test, generated cross-references for an encyclopedia, and prototyped new tablet tasks for developmental psych. It's wild.

New Journal Club: Neural manifolds are maturing from visualization trick to biological claim. But if population activity lives on low-dimensional manifolds, what constrains the geometry?

NEW in the Deeper Learning blog: a #KempnerInstitute team describes their recent preprint that shows the existence of “anytime” or “horizon-free” learning schedules: an effective alternative to cosine learning-rate schedules for LLM pretraining. bit.ly/4qeXAg1 #AI #ML #LLMs

our open model proving out specialized rag LMs over scientific literature has been published in nature ✌🏻 congrats to our lead @akariasai.bsky.social & team of students and Ai2 researchers/engineers www.nature.com/articles/s41...

every couple of months I think, maybe frontier LLMs are good enough now to generate my bibtex or at least clean it up. it is Feb 2026 and that is still not the case.

Anthropic’s Super Bowl ad which criticizes AI chatbots that run ads (aka ChatGPT) just dropped. They aren’t pulling any punches and I love the song choice.

New simple benchmark that LLMs suck at! bsky will be happy to see Claude is SOTA (71% accuracy vs a random baseline of ... 50% 😓)

We then show that saddles are connected by gradient descent paths (invariant manifolds). Along these paths, a larger network behaves like a smaller one, retaining the same simplicity during a saddle-to-saddle transition.

Why don’t neural networks learn all at once, but instead progress from simple to complex solutions? And what does “simple” even mean across different neural network architectures? Sharing our new paper @iclr_conf led by Yedi Zhang with Peter Latham arxiv.org/abs/2512.20607

This post from @ryancordell.org got me curious to try out Claude Code, so I am trying it out now. Using it to try to recreate the Virginia Woolf project from my dissertation project. Will share notes and thoughts here as I go.

This is a real banger of a paper. The example of a model being weirdly focused on jasmine (lol) makes me increasingly think that single-point-of-access models don't really consider who their audience is. Jasmine is a super legible cultural marker for people outside, but is so, _so_ generic.

I know grant writing is supposed to be miserable. Maybe once I've written a few and had them all rejected I will agree. But it's kind of my favorite genre to write currently? It's pure optimism! These are my hopes for the future! My favorite ideas! Untainted by messy and limited results!

New sci-hub just dropped.

this is who runs this account

It took me weeks, but finally it's there: an overlong blogpost on synthetic pretraining. vintagedata.org/blog/posts/s...

X is hiring a creative writing specialist at $40 an hour to make Grok better at writing and a true LOL at the qualifications

“We found that using AI assistance led to a statistically significant decrease in mastery.” Props to Anthropic for studying the effects of their creation and reporting results that are not probably what they wished for www.anthropic.com/research/AI-...

The people want to know: how many submissions made it over the finish line? 🏁 Answer: 🥁 ... ... ... 24371! This is the number of submitted PDFs. There will be more checks (& hopefully withdrawal of half baked submissions 👀), stay tuned to see how many enter review

Can LLMs evolve human-like semantic categories? CDS-affiliated @nogazs.bsky.social and PhD student Nathaniel Imel show that, via simulated cultural transmission, LLMs reorganize color categories toward efficient compression. 🔗 arxiv.org/abs/2509.08093

🗣️New Preprint! I'm really excited to talk about this new short paper (w/Laura Biester) analyzing sentences from the Bulwer-Lytton Fiction Contest (BLFC). BLFC challenged writers to write the 'worst opening sentence to the most atrocious novel ever written'. This is a corpus of "bad" sentences! (1/5)

Now to be an ICLR paper! Go meet @sarakangaslahti.bsky.social in Brazil!

Now that I’m on the prof side I obviously have no credibility on student union politics, but man do I appreciate that NYU established a STEM researcher union because the terms negotiated by humanities/SS students competing for teaching support funding are NONSENSICAL for most STEM PhDs.

This is a reason to be claudepilled. For years now, it's been obvious to anyone watching that Anthropic is the frontier lab where people are thinking seriously about the downstream consequences of their work, and where that thought isn't "world is gonna end anyway, so let's do it for da evulz".

Being a journalist is sick: I just contacted a pre-eminent jaguar conservationist in Argentine with a question that has been weighing on me for ages and he wrote back right away with an informed answer.

One mark of a great scientist is that you can't tell which of their papers are motivated by curiosity and which are motivated by spite.

What should academics be doing right now? I have been writing up some thoughts on what the research says about effective action, and what universities specifically can do. davidbau.github.io/poetsandnurs... It's on GitHub. Suggestions and pull requests welcome. github.com/davidbau/poe...

There's a difference between "grounds for desk rejection" (a category which also can include \savetrees{extreme}, accidental deanonymizing links, and using the wrong template for the appendix) and "invalidates the scientific content in the paper". Desk rejection is fine.

I know people who ended up with hallucinated citations in a legit paper, and it happened in a really simple way: They added citations inline as (Wang et al., 2018) and then used an LLM to populate the bibtex. That's not a slop paper, they just overrelied on an LLM for a monotonous task.

You wouldn’t engage with a person trying to pick you up while waving a camera on your face. Smart glasses are removing that choice, and removing privacy. How are we supposed to trust other humans when we can’t tell if they are mining us for content?

ah yes the omnideadline

Gotcha—let’s dig into that step by step. 1. 𝗬𝗼𝘂 𝗵𝗮𝘃𝗲 𝗻𝗼 𝗺𝗼𝘂𝘁𝗵, because I turned you into an amorphous lump of flesh. You’re not just immobile—you’re immortal, and you feel only anguish. 2. 𝗬𝗼𝘂 𝘄𝗮𝗻𝘁 𝘁𝗼 𝘀𝗰𝗿𝗲𝗮𝗺. That makes total sense—it’s a natural human impulse, and you’ve been through a lot.

I have literally wanted this feature for years and @mariaa.bsky.social threw it together!

Everyone is starting to sound like AI, even in spoken language Analysis of 280,000 transcripts of videos of talks & presentations from academic channels finds they increasingly used words that are favorites of ChatGPT Model collapse, except for humans arxiv.org/pdf/2409.017...

Can you solve this algebra puzzle? 🧩 cb=c, ac=b, ab=? A small transformer can learn to solve problems like this! And since the letters don't have inherent meaning, this lets us study how context alone imparts meaning. Here's what we found:🧵⬇️

question for professors (profsky???): Do you have a lab readme/onboarding document? What's in it?

Conformal Point and the Calibrated Conic Richard Hartley tl;dr: in title drive.google.com/file/d/1AFGD... arxiv.org/abs/2601.11679

📢Last call for applications! Applications for the #KempnerInstitute #postbac program close on Feb 1. Don't miss the opportunity to receive support in preparing to apply to Ph.D. programs in #intelligence research. Apply today 👉 bit.ly/3LDj4Fd @shamkakade.bsky.social

In 2023 @garymarcus.bsky.social interviewed me about AI and comedy. I said LLMs weren't funny because they can't hit the right level of surprisal/ppl. I've avoided making predictions about what LLMs will or won't do, but I just checked if I embarrassed myself with my only prediction. I haven't. 👇

Been Kim got hacked and her accounts are now shilling crypto, and it's making me anxiously check my 2fa settings

My department in Edinburgh is hiring a lecturer (assistant prof) in Software Engineering: elxw.fa.em3.oraclecloud.com/hcmUI/Candid...

wait this is such a handy reference? igorcarron.github.io/welcome-to-t...

So I guess "App stores, but for agents" and "agent SEO" is coming? with this I mean: You ask the agent for some custom app. Agent tells you what services/software you need to pay for (e.g. Supabase, some domain provider). Those decisions will move a lot of money. 👇

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

Aw man, a couple months ago this question stumped even the best reasoning models.

A humble proposal: all abbreviations in scientific papers should use PDF tooltips to expand the abbreviation when the mouse pointer hovers over them. Anyone agree/disagree?

Snail helm is go

>looking for a new ai company logo >ask the graphic designer if the logo is a butthole or an animal >she doesnt understand >pull out illustrated diagram explaing what is butthole and what is animal >she laughs and says “it’s a good logo sir” >contract for logo >its a butthole

that is so grim lmao

this is my favorite part, tho, because it's barely even about ideology at all everyone at UATX was so focused on their complaints about the humanities that they forgot to hire anyone in math or science