Kim Stachenfeld, PhD

We Will Not Be Divided Employees of Google and OpenAI stand together to refuse the Department of War's demands to use AI models for domestic mass surveillance and autonomous killing without human oversight.

Pleased to see some friends' names here :)
notdivided.org

Updated work from @jessegeerts.bsky.social extending his results on transitive inference in transformers (including LLMs!)

updated paper: arxiv.org/abs/2506.04289
bleeprint (what are we calling these?) below ⬇️

How neuroscientists are using AI Eight researchers explain how they are using large language models to analyze the literature, brainstorm hypotheses and interact with complex datasets.

Researchers are using LLMs to analyze the literature, brainstorm hypotheses, build models and interact with complex datasets. Hear from @mschrimpf.bsky.social, @neurokim.bsky.social, @jeremymagland.bsky.social, @profdata.bsky.social and others.

#neuroskyence

www.thetransmitter.org/machine-lear...

hahah brings me back!

Our summer program values mentorship as much as scientific skill-building, connecting Africa’s talents with experts in both #academia and #industry. Grateful to @neurokim.bsky.social & @ankahira.bsky.social for inspiring our 2025 cohort. @columbiauniversity.bsky.social @astra-zeneca.bsky.social

🧠 How do transformers learn relational reasoning? We trained small transformers on transitive inference (if A>B and B>C, then A>C) and discovered striking differences between learning paradigms. Our latest work reveals when and why AI systems generalize beyond training data 🤖

New work on relational reasoning in transformers!

TLDR: Inductive biases of In-Weight and In-Context Learning in transformers are really different for relational reasoning, and pretraining can make a big difference for in-context.

Check out @jessegeerts.bsky.social's thread for more!

This is fucking insane

💥 Do we now have all the components necessary to reach #AGI, or are there more discoveries to be made? Can the study of the brain help provide them?

We tackle these questions and more at our Conference on the Mathematics of Neuroscience and AI in sun-soaked Split, Croatia:

www.neuromonster.org

Tomorrow is our @cosynemeeting.bsky.social workshop on Graph Neural Networks! From molecules to circuits to biomechanics to reasoning behaviors to social behaviors, neuroscience is rich with geometrically structured data.

Our goal: demystify GNNs, talk opportunities for 🧠

@cosynemeeting.bsky.social workshops are almost here! Join GNN workshop w @neurokim.bsky.social and Sam Lewallen to hear @herrsaalfeld.bsky.social head of computation @hhmijanelia.bsky.social present on GNNs for learning structure and function underlying neural assemblies #cosyne2025

Her lab does impressive foundational work on geometric statistics and manifold learning, sure to be an amazing talk! Come for the GNNs, stay for the manifolds! #cosyne2025 @neurokim.bsky.social

More 🔥🔥 speakers at @cosynemeeting.bsky.social GNN workshop: @ninamiolane.bsky.social who runs the @geometric-intel.bsky.social lab at UCSB. She will take us beyond 🚀🚀 GNNs, presenting a survey of message passing topological neural networks for neuroscience

Rounding out our speakers @cosynemeeting.bsky.social GNN workshop is @dom-beaini.bsky.social of @mila-quebec.bsky.social and @valenceai.bsky.social. Come hear him tell us how to learn a…MOLECULE! Don’t forget brains 🧠 are made of chemicals ⚗️ and chemicals + graphs = 🫶

Sign the Petition Free Mahmoud Khalil. Stop His Unjust Detention.

Sign these petitions to release detained students:
www.change.org/p/free-mahmo...

www.change.org/p/free-rumey...

www.change.org/p/release-un...

@change.org

If folks know of opportunities to do this, I'd love to hear about them (DM me). This isn't something I (...like lots of us) have experience in maybe, but no time like the present.

Trump officials target dozens of colleges for antisemitism - even those cleared by probes When Muhlenberg College learned in January 2024 - as pro-Palestinian protests were sweeping campuses across the United States - that it was under investigation for discriminating against Jewish students, the small liberal arts institution in Pennsylvania took action.

Universities need to fight back with what we've got, which is not nothing: legal, comms, organizing... Plus, there are a whole bunch of us! 60 universities on this list, all with a common economic interest (not to mention a moral one).
www.reuters.com/world/us/the...

This is not just happening at Columbia. This is happening all over, to students, who have espoused a political opinion. Free speech, due process, intellectual freedom... we need that for stuff.

Tufts student had asthma attack after ICE arrest, attorneys say Lawyers for Rumeysa Ozturk demand her release after ‘shocking, cruel and unconstitutional’ arrest

Rumeysa was sent from Massachusetts to Louisiana -- just like Mahmoud Khalil -- despite a court blocking her removal from the state. She had an asthma attack while being sent to Louisiana.
www.independent.co.uk/news/world/a...

Surveillance shows Tufts graduate student detained YouTube video by WCVB Channel 5 Boston

There is footage of her arrest by plain clothes officers. I cannot imagine how terrifying it must be to have strangers walk up to you on your way to dinner and put you in a car and you're gone.
www.youtube.com/watch?v=PuFI...

Tufts community calls the Tufts administration to action We, members of the Tufts community, are appalled and enraged by the unlawful abduction on Tufts PhD Fulbright student Rümeysa Öztürk by federal authorities. We call on the Tufts administration to comm...

If you are a member of the Tufts community, there is a petition to encourage the Tufts administration to take a more active role in protecting and supporting at-risk students.
docs.google.com/forms/d/e/1F...

Beautiful piece about Rumeysa Öztürk, the tufts student who was abducted and detained for political speech, which paints a picture of a warm, compassionate, beloved member of the Tufts community (my alma mater) from those who know her.

as.tufts.edu/epcshd/news-...

Just over a week to #cosyne2025 workshops! Speaker profile number 2 for GNN workshop w/ @neurokim.bsky.social and Sam Lewallen is @kristinmbranson.bsky.social of @hhmijanelia.bsky.social!

Kristin’s lab develops cutting edge machine vision approaches for quantitatively studying animal behavior.

ARIA Opportunity Space: Scalable Neural Interfaces

Looking for an exciting fellowship in AI & Neuro, with competitive salary (~£100k)? We got a new position in the lab at Oxford, working with @somnirons.bsky.social and me! 🧪

Our project: encode.pillar.vc/projects/beh...

General info: encode.pillar.vc

#compneuro #neuroai #neuroscience #sciencejobs

Finally this thread was shamelessly plagiarized from @kennethmarino.bsky.social 's thread on X (thanks kenny!):

x.com/Kenneth_Mari...

Thanks to colleagues @ndrewliu.bsky.social, Henry Prior, Gargi Balasubramanian, Rivka Moroshko, Amir Zait, Ilia Labzovsky, Danny Karmon, Ishita Dasgupta, and @kennethmarino.bsky.social at Google DeepMind for a fun and super interdisciplinary collab!! 🧠🤖

For more information, check out the paper
openreview.net/pdf?id=yORSk...

and the open-source code + tutorial, where you can make custom dataset generators.
github.com/google-deepm...

Talk to @kennethmarino.bsky.social at #ICLR2025 in Singapore if you want to learn more!

Third, it lets us generate custom datasets and configurations which can be used, for instance, compare reasoning behaviors across humans and models.

This also can be useful for experimental psychologists for generating new experiments in humans, and piloting experiments with LLMs.

Second, it lets us more easily probe and diagnose models. Because of our configurability and synthetic generation, we can make intentional changes and ablations to the data to better diagnose model issues as well as generate auxiliary labels for analysis.

This approach gives us three big advantages.

First, it lets us scale the amount of context and (importantly) the complexity of the graph problem to the capability of the LLM letting our problems scale as models become increasingly long-context and powerful.