How is uncertainty in LLMs output reflected in internal representations? In our new work (to appear at ICML 2026), we show that the shape of internal token trajectories provides a direct geometric link to behavioral uncertainty (output entropy). 🧵(1/n)

Our piece on language-like properties in animal vocal behavior is available today. Check it out! This is an incredibly difficult and thorny topic which is nonetheless of supreme importance and interest imco (in my current opinion)

To accompany my textbook (Computational Foundations of Cognitive Neuroscience) and the class I taught this semester, I'm open-sourcing my lectures slides: gershmanlab.com/lectures.html I'll continue to update these as I improve them.

When and how can test-time thinking allow models to use information latent in their training data? What are the benefits and tradeoffs relative to other solutions like synthetic data augmentation? Pleased to share (after a long delay) an exploration of these issues: arxiv.org/abs/2604.01430 thread:

Pleased to share that our paper "Representation Biases: Variance is Not Always a Good Proxy for Importance" is now out as Theory/New Concepts paper in eNeuro! www.eneuro.org/content/13/3... 1/

Excited to share these two works at @cosynemeeting.bsky.social ! On Friday, Jack King presents his work on "Representational curvature shapes uncertainty in large language models." 📍 Poster 2-223 Done in collaboration with @evfedorenko.bsky.social. 🧵👇

New paper! In arxiv.org/abs/2601.20834 we study how language models representations of things like factuality evolve over a conversation. We find that in edge case conversations, e.g. about model consciousness or delusional content, model representations can change dramatically! 1/

The visual world is composed of objects, and those objects are composed of features. But do VLMs exploit this compositional structure when processing multi-object scenes? In our 🆒🆕 #ICLR2026 paper, we find they do – via emergent symbolic mechanisms for visual binding. 🧵👇

How do diverse context structures reshape representations in LLMs? In our new work, we explore this via representational straightening. We found LLMs are like a Swiss Army knife: they select different computational mechanisms reflected in different representational structures. 1/

Really excited about this work by @lampinen.bsky.social and the team! Being a small part of this work, I learned a lot about representation drift during in-context learning, role-play, and the challenges of a static view on interpretability. Highly recommend it :)

The wait is over! I am so excited to start over and re-learn the foundations from @gershbrain.bsky.social :)

Presenting this afternoon from 1-5pm!

The UniReps Workshop accepted papers are out! 🎉 Huge thanks to the authors, reviewers, and incredible AC committee for their dedication and effort in making this happen. 🙏 openreview.net/group?id=Neu...

🚨I am looking for a POSTDOC, LAB MANAGER/TECH and GRAD STUDENTS to join my new lab in beautiful Madison, WI. We study how our brains perceive and represent the physical world around us using behavioral, computational, and neuroimaging methods. paulunlab.psych.wisc.edu #VisionScience #NeuroSkyence

This is an amazing opportunity: work with Vivian Paulun ‪@vivianpaulun.bsky.social‬, who has deep insights about perception, and extraordinary skills as an experimentalist and theorist.

LLMs get a lot of attention but if you are interested in the original (allegedly) stochastic parrots come to my poster Tuesday where I will present new work on the natural vocal behavior or parrots and its cortical control #Sfn2025 #SfN25

Wow what a nice picture I wonder who took it :)

Neuronal computation in the cerebellum via a vector calculus. Work of Mohammad Amin Fakharian, Alden Shoup, Paul Hage, and Hisham Elseweifi www.science.org/doi/10.1126/...

A detailed schedule for CCN2025 is available on our website now, including the specific keynotes / GACs / community events taking place: 2025.ccneuro.org/schedule-of-... The early bird registration deadline is coming up this Friday (23rd May)!

New paper! 🧠 **The cerebellar components of the human language network** with: @hsmall.bsky.social @moshepoliak.bsky.social @gretatuckute.bsky.social @benlipkin.bsky.social @awolna.bsky.social @aniladmello.bsky.social and @evfedorenko.bsky.social www.biorxiv.org/content/10.1... 1/n 🧵

📣 Grad students and postdocs in computational and theoretical neuroscience: please consider applying for the 2025 Flatiron Institute Junior Theoretical Neuroscience Workshop! All expenses are covered. Apply by April 14. jtnworkshop2025.flatironinstitute.org

I couldn’t recommend a better place to learn and work if you’re considering neuroscience grad school and want to interact with phenomenal scientists!

So excited to have our work on conlangs out in PNAS: www.pnas.org/doi/10.1073/... Congrats, Saima, Maya, and the rest of the crew -- well done! Here is the MIT news story: news.mit.edu/2025/esperan...

Very important work by @rarefin.bsky.social, Oscar Skean, @yann-lecun.bsky.social and @shwartzzzivravid.bsky.social on richness of information content in intermediate layers in LLMs. highly recommended

🚨 First preprint from the lab! 🚨 Josh Rozner (w/@weissweiler.bsky.social and @kmahowald.bsky.social) uses counterfactual experiments on LMs to show that word distributions can provide a learning signal for diverse syntactic constructions, including some hard cases.

What should count as a good model of intelligence? AI is advancing rapidly, but how do we know if it captures intelligence in a scientifically meaningful way? We propose the *NeuroAI Turing Test*—a benchmark that evaluates models based on both behavior and internal representations.

Many neuroscience students are steeped in an experiment-first style of thinking that leads to “random walk science.” Let’s not forget how theory can guide experiments towards deeper insights, writes @gershbrain.bsky.social. #neuroskyence www.thetransmitter.org/theoretical-...

Nature research paper: Children’s arithmetic skills do not transfer between applied and academic mathematics https://go.nature.com/40PT3pT

📢 Applications are open for 2025 Analytical Connectionism! Topical focus: bias in learning Learn key connectionist models & analytical tools for neural-network analysis, and work on novel projects with faculty & postdoc mentors ⏰ Apply by 18 April ℹ️ www.analytical-connectionism.net/school/2025

Our representational alignment workshop returns to #ICLR2025! Submit your work on how ML/cogsci/neuro systems represent the world & what shapes these representations 💭🧠🤖 w/ @thisismyhat.bsky.social @dotadotadota.bsky.social, @sucholutsky.bsky.social @lukasmut.bsky.social @siddsuresh97.bsky.social

Check our latest in which we leverage shape metrics to compare neural geometry across regions, sessions or subjects and how their differences predict behavior. w/ Nejatbakhsh, Duong, @sarah-harvey.bsky.social, Brincat, @siegellab.bsky.social, @earlkmiller.bsky.social & @itsneuronal.bsky.social

Why do diverse ANNs resemble brain representations? Check out our new paper with Colton Casto, @nogazs.bsky.social , Colin Conwell, Mark Richardson, & @evfedorenko.bsky.social on “Universality of representation in biological and artificial neural networks.” 🧠🤖 tinyurl.com/yckndmjt

New paper from our lab, by Mark Saddler, using machine learning to test the role of temporal coding in hearing. Here is a quick summary. (1/n) www.nature.com/articles/s41...