Excited to announce CCN @CogCompNeuro Satellite Event: Modeling and Understanding Human Brain Computation at Scale We explore how we can best leverage the many recent advances with neural network modeling and brain foundation models toward theoretical insight and benefit for humanity. Sunday, August 2, 2026 @ZuckermanBrain Organized with @Pinyuan3, @LibraCheng, Andrew Luo, and @KriegeskorteLab

What can a neuron compute? Real biological neurons are complex, but how capable are they? Using a new method, we found that a single cortical neuron can classify cats vs dogs, recognize spoken words, and solve 10-bit parity, all tasks thought to require entire networks. (1/15)

Why do humans generalize so much better than deep networks? Because they learn something deep networks can't: symmetries. arxiv.org/abs/2412.11521

gm! aunties room 01 clutter, housework, sisters.

We introduce a method for training RNNs that is time-parallel and does not suffer from vanishing/exploding gradients. Key idea is to decouple learning 1) what should be remembered (can be done without recurrence) and 2) how to update memory (can be one-step supervised by #1).

“A call to abandon the conventional boundaries between computer vision and robot learning, and instead ponder the problems that arise when we seek to build machines that both perceive and act.”

Can AI agents really peer-review Nature papers? 🔬🤖 Moving beyond simple rating consistency, we conducted a large-scale study analyzing how well AI vs. humans point out paper limitations across Nature/Nature sister journals. We found that with a carefully designed harness, AI is supremely effective, though it lacks the diverse perspectives of humans. Check out @seungonekim ‘s thread & our 91-page comprehensive study!

Can AI "see" motion from static images like humans? We tested 10 state-of-the-art vision models (multi-scale, recurrent, and bio-inspired) on the Rotating Snakes illusion, one of the most extensively characterized motion illusions in human and primate vision. Almost all failed — only one bio-inspired model, Dual, predicted motion most closely aligned with human perception, particularly under simulated eye movements. Accepted to @CVPR Findings! Paper: openaccess.thecvf.com/conten… w/ @LibraCheng @zitangsun @KriegeskorteLab A 🧵: #ComputerVision #Vision

You see rotation when you move your eyes, even though the image is 100% static. Our #CVPR Findings paper uses motion illusions to (1) reveal gaps between human and AI motion processing and (2) identify architectures key for aligning vision models with human perception. openaccess.thecvf.com/conten…

Every post on X is now an ad: visibility depends on payment (subscription tier, boosts). Twitter/X was never perfect, but has recently been moving decisively away from a conception that is compatible with the ideals of liberal democratic discourse. [Advertisement]

Rich Sutton saying that creativity requires the cycle of variation, evaluation, and selective retention as present in RL and known as instrumental conditioning in psychology. He argues that, absent this cycle, a system will never learn to create something that is both novel and good. No combination of supervised learning by backprop, prediction, world modeling can lend generative AI creativity. It's a pretty strong claim. youtu.be/K5LAFEjTlBA

1/ New preprint VSS poster! Attention has long been thought to enable efficient vision. But does it? First demonstration that attention—consuming just 4-5% of the energy budget—can cut the energetic costs of vision in half. #VSS2026

Look. The snakes rotate. Hold still. They stop. Glance away. They spin again. We asked 10 visual-motion models to play this game with us. Only one could. 🐍 I'll present "Neural networks reveal candidate computational mechanisms underlying anomalous motion illusion" at #VSS2026 🗓️ May 19th, Tuesday 8:30–12:30 📍 Pavilion, Board 53.435 Welcome to drop by! 👋 w/ @IsabellaRosario @ZitangSun @KriegeskorteLab

General intelligence is generic. Domain specific intelligence is more interesting.

Want a digital computer to write symbolic programs for itself? Simulate on it a subsymbolic mind & brain somewhat like ours: a sea of associations continually transformed. Symbolic thought, who knew, comes more naturally to a neural network than to a symbolic architecture.

A service “chat” on Uber and Amazon means the customer reads through a list of issues other customers had in the past and picks one. The stupidity of this, in the age of AI chat models, boggles the mind — unless the goal is not to serve the customer.

- Drafted a blog post - Used an LLM to meticulously improve the argument over 4 hours. - Wow, feeling great, it’s so convincing! - Fun idea let’s ask it to argue the opposite. - LLM demolishes the entire argument and convinces me that the opposite is in fact true. - lol The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.

Today we present a new framework for measuring human-like general intelligence in machines (what some people call AGI). Conventional AI benchmarks today assess only narrow capabilities in a limited range of human activities. We propose that a more promising way to evaluate human-like general intelligence in AI systems is through a particularly strong form of general game playing: studying how and how well they play and learn to play all conceivable human games — what we call the ``Multiverse of Human Games''. Taking a first step towards this vision, we introduce the AI GameStore, a scalable and open-ended platform that uses LLMs with humans-in-the-loop to automatically construct standardized and containerized variants of popular human games on digital gaming platforms. As a proof of concept, we generated 100 such games based on the top charts of Apple App Store and Steam, and evaluated seven frontier vision-language models (VLMs) on short episodes of play. The best models achieved less than 10% of the human average score on the majority of the games. Check out our website to play the games, see how agents play, and build agents to solve them!

Nice to see new work exploring this direction of "going beyond linear transformers with proper recurrence"! For context, see our related work from NeurIPS 2021: arxiv.org/abs/2106.06295, a follow-up to our ICML 2021 DeltaNet paper. w/ @ImanolSchlag @robert_csordas @SchmidhuberAI