Alireza Karami | علیرضا کرمی

We recently stumbled upon a surprisingly common misunderstanding in computing noise ceilings that can be quite consequential. So if you care about noise ceilings, please check out Sander’s thread and our preprint! 👇

Left: washing machine with the words “Critical Washing” on it. Right: Quote: “When we critique AI, we should do so with intellectual honesty and in a principled way. Utmost care is needed to avoid ethics washing, greenwashing, and generally — what we dub — critical washing. (…) Researching and reflecting on the harms of AI is not itself harm reduction. It may even contribute to rationalizing, normalizing, and enabling harm. Critical reflection without appropriate action is thus quintessentially critical washing.” —Suarez, Müller, Guest, & van Rooij (2025) https://zenodo.org/records/15677840

New today in @Nature: your visual cortex contains touch-based body maps. bit.ly/VisualBodyMaps
Your brain transforms what you see into first-person, body-referenced codes: A previously unknown bridge between vision and touch.

New work with @simyciri.bsky.social. Adults make more accurate decisions than adolescents—yet show more intrusion from irrelevant players. We explain this paradox via efficient information compression: adults rely on combinatorial social basis functions that use group structure as a mental scaffold.

Y’all are reading this paper in the wrong way.

We love to trash dominant hypothesis, but we need to look for evidence against the manifold hypothesis elsewhere:

This elegant work doesn't show neural dynamics are high D, nor that we should stop using PCA

It’s quite the opposite!

(thread)

Émission frustrante. Soyons clairs: La psychanalyse :
- ne fait pas progresser les enfants
- s’appuie sur un fatras de chimères théoriques (œdipe, castration, jeux de mots Lacaniens) qu’on a encore entendues ce soir
- conduit trop de CMPP à culpabiliser les parents au lieu de les outiller

Numerosity Is Directly Sensed and Dynamically Transformed in the Human Brain: Evidence from MEG-MRI Fusion https://www.biorxiv.org/content/10.1101/2025.11.15.687894v1

Noise ceilings are really useful: You can estimate the reliability of your data and get an index of how well your model can possibly perform given the noise in the data.

But, contrary to what you may think, noise ceilings do not provide an absolute index of data quality.

Let's dive into why. 🧵

AI and Neuroscience | IVADO

I’m looking for interns to join our lab for a project on foundation models in neuroscience.

Funded by @ivado.bsky.social and in collaboration with the IVADO regroupement 1 (AI and Neuroscience: ivado.ca/en/regroupem...).

Interested? See the details in the comments. (1/3)

🧠🤖

I think “Academia = papers” vs. “Science = helping society” is too simplistic. Science isn’t always about direct utility; sometimes its value is understanding itself. And academia isn’t defined by papers — that’s just the current incentive system.

An exciting opportunity to work with @martinhebart.bsky.social

Pleased to share new work with @sflippl.bsky.social @eberleoliver.bsky.social @thomasmcgee.bsky.social & undergrad interns at Institute for Pure and Applied Mathematics, UCLA.

Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
www.arxiv.org/pdf/2510.15987

🧵1/n

Could these “function vectors” in LLMs resemble the neural representations in the human brain when people do similar tasks? Has anyone tested this with fMRI or MEG data?

When does new learning interfere with existing knowledge in people and ANNs? Great to have this out today in @nathumbehav.nature.com

Work with @summerfieldlab.bsky.social, @tsonj.bsky.social, Lukas Braun and Jan Grohn
www.nature.com/articles/s41...

For your enjoyment (by @jagarikin)

From ancient Greece to the Arabic golden age, scholars have been driven by their curiosity to investigate astronomy, history, philosophy, and sundry other disciplines. Is there a structure to that curiosity? Are astronomers as likely to also be historians or to also be philosophers?

📣 Apply now for the 🌵CaCTüS 3-month fully funded internship in Germany in 2026!
Work on ML, theor. neuroscience, computational psychiatry, behavioral experiments or data analysis at a top institution in Tübingen. For students facing barriers in higher education. 👉 cactus-internship.tuebingen.mpg.de

Ten Simple Rules for AI-Assisted Coding in Science While AI coding tools have demonstrated potential to accelerate software development, their use in scientific computing raises critical questions about code quality and scientific validity. In this pa...

Ten Simple Rules for AI-Assisted Coding in Science arxiv.org/abs/2510.22254 - our latest, led by @ericwbridgeford.bsky.social

Semantic Tuning of Single Neurons in the Human Medial Temporal Lobe The Medial Temporal Lobe (MTL) is key to human cognition, supporting memory, emotional processing, navigation, and semantic coding. Rare direct human MTL recordings revealed concept cells, which were ...

🚨Preprint: Semantic Tuning of Single Neurons in the Human Medial Temporal Lobe

1/8: How do human neurons encode meaning?
In this work, led by Katharina Karkowski, we recorded hundreds of human MTL neurons to study semantic coding in the human brain:

doi.org/10.1101/2025...

I am close to tears; they listened and made an english version apache.be/2025/10/24/b...

@apache.be you rule 🖤

How does our brain excel at complex object recognition, yet get fooled by simple illusory contours? What unifying principle governs all Gestalt laws of perceptual organization?

We may have an answer: integration of learned priors through feedback. New paper with @kenmiller.bsky.social! 🧵

Please enjoy my series of essays on the subtle distinction between the game-theoretic concepts of tactics and strategies.
www.infinitelymore.xyz/t/tactic #InfinitelyMore

Interesting PhD positions in AI & ML with the ELLIS–Max Planck AI Network. Potential opportunities for interdisciplinary projects bridging AI and neuroscience.

Variance partitioning is used to quantify the overlap of two models. Over the years, I have found that this can be a very confusing and misleading concept. So we finally we decided to write a short blog to explain why.
@martinhebart.bsky.social @gallantlab.org
diedrichsenlab.org/BrainDataSci...

I wanted to add some thoughts to this excellent blog post, not detailed, maybe wrong, maybe useful:
1. Unique variance is easy to interpret as a lower bound of what a variable explains (the upper bound being either what the variable explains alone or what the other variables cannot explain uniquely)

Dissecting LLMs with ML | Mike X Cohen | Substack Understand how large language models (LLMs) really work by applying machine learning (ML) methods to their internal activations. Each post explores how LLMs process text, isolate patterns, and generat...

Dear scientists of BlueSky: Wouldn't it be great to learn how large language models work by running experiments on them and analyzing the data? Yeah, you can do that, and I'll show you how.
mikexcohen.substack.com/s/ml-on-llms

I’m recruiting committee members for the Technical Program Committee at #CCN2026.

Please apply if you want to help make submission, review & selection of contributed work (Extended Abstracts & Proceedings) more useful for everyone! 🌐

Helps to have: programming/communications/editorial experience.

Thrilled that #CCN2026 will be hosted at NYU! Having experienced an amazing summer school there last year, I know how inspiring the city is. It’s heartbreaking, though, that many brilliant Iranian students and researchers may be excluded due to the new travel restrictions.

Come and check out our poster at #CCN2025, presented by @tlmnhut.bsky.social

Investigating action topography in visual cortex and deep artificial neural networks High-level visual cortex contains category-selective areas embedded within larger-scale topographic maps like animacy and real-world size. Here, we propose action as a key organizing factor shaping vi...

New preprint out! We propose that action is a key dimension shaping the topographic organization of object categories in lateral occipitotemporal cortex (LOTC)—and test whether standard and topographic neural networks capture this pattern. A thread:

www.biorxiv.org/content/10.1...

🧵 1/n