Patrick Mineault

Excited to release what we’ve been working on at Amaranth Foundation, our latest whitepaper, NeuroAI for AI safety! A detailed, ambitious roadmap for how neuroscience research can help build safer AI systems while accelerating both virtual neuroscience and neurotech. 1/N

Looking for an even more considered take on the state of modern systems neuroscience? Read my piece in Macroscience, IFP’s fantastic (meta)science newsletter! I explain why systems neuroscience is like the Matrix, delve more into state-of-the-art neuroscience tools, and explain why sustained team science efforts are essential if we someday hope to upload our brains to the cloud. macroscience.org/p/how-close-ar…

Excited to share my postdoc work is out in @NatureNeuro today! We examined how the brain enables social groups to collectively coordinate their behavior in the face of environmental challenge ❄️🐭🐭🐭🐭❄️ : nature.com/articles/s4159…

⭐️ New preprint — work led by @mdunnhofer , supported by Jean de Dieu U. biorxiv.org/content/10.648… In natural scenes, objects are often hidden by clutter, occlusion, or camouflage. Yet humans can suddenly “see” them the moment they move. To study this, we leveraged the 𝗠𝗢𝗖𝗔 (Moving Camouflaged Animals) dataset — a unique set of videos where objects are nearly invisible in static frames but become perceptually clear through motion. In our new work, we asked: 👉 Do modern AI systems rely on motion the same way we do? We brought together: • 🧑🏽‍💻👨🏽‍💻Human behavior • 🐒🧠Macaque IT neural recordings • 🤖Image-based and video-based neural networks Here’s the surprising part: Image-based neural networks are already very good at estimating object properties — even in these highly challenging, camouflaged scenes from MOCA. But that’s exactly the problem. Because they perform so well using static cues alone, they don’t need to rely on motion the way biological systems do. As a result, they fail to show the strong motion-dependent improvements seen in humans and the primate brain. Video models partially recover this behavior by integrating information over time — but still fall short of fully matching biological vision. The takeaway: Better performance does not mean better models of the brain. In fact, being “too good” at static recognition may push models toward the wrong computational strategies. Datasets like MOCA expose this gap clearly: 👉 Humans sometimes need motion to see. 👉 Models often don’t. If (we might not) we want AI systems that truly reflect how primate vision works, we need to go beyond static benchmarks and capture the dynamic computations that stabilize perception in the real world.

New paper! Deep-learning-assisted simulation of a cortical circuit: integrating anatomy, physiology and function biorxiv.org/cgi/content/sh… It comes with a new major model release from @AllenInstitute. Find models and code here: github.com/AllenInstitute… 🧵 1/13

We know about cosmological dark matter despite being unable to measure it because, without it, galaxies would fall apart. By analogy, let's talk about "cognitive dark matter" (CDM): brain functions that meaningfully shape behavior but are hard to infer from behavior alone. New paper! 🧵👇

Very neat work on the compute-only path to intelligence:

Sharing “Neural Thickets”. We find: In large models, the neighborhood around pretrained weights can become dense with task-improving solutions. In this regime, post-training can be easy; even random guessing works Paper: arxiv.org/abs/2603.12228 Web: thickets.mit.edu 1/

@HistedLab @CosyneMeeting @neuro_kim Jealous!

Fabulous keynote session at @cosynemeeting - touching on neuroscience, AI, and mech interp. @neuro_kim and I kicked off the meeting and introduced Chris Olah. 1/

Our live tissue clearing paper is out in @naturemethods! We achieved optical clearing of mammalian brain tissues without compromising normal neuronal function. Big congrats to @Shigenori774 and our wonderful collaborators! 🎉 nature.com/articles/s4159… (1/10)

There's no better way to learn than to teach! Help make NMA a resounding success!

Thanks @patrickmineault for sharing our work with @barabasi in your NeuroAI series on cell types and connectomes! An excellent summary on how much info is unlocked in development, and how these processes can be specific enough to hardwire innate behaviors.

So proud of what we've built in neuroAI and longevity at Amaranth over the last few years! I'm excited that we're unveiling that vision of the future—expect more essays from James and the team on frontier tech (including some from me!)

Excellent Substack writeup (substack.com/home/post/p-18…) by @patrickmineault on how cell types may specify innate behaviors and why mapping regions of the brain specialized to steer innate behaviors (via lots of distinct cell types) could lead us to more aligned AI systems.

@patrickmineault explains what I struggled to on the cell types connection. Love this. Field building this area a bit recently we now have (in addition to Zador, Barabási and other papers on e.g. genetic encoding of connectomes)… Expository essay: asteriskmag.com/issues/13/the-… Byrnes core theory base: zenodo.org/records/179535…, osf.io/preprints/osf/… Steering subsystem cell types preprint: preprints.org/manuscript/202… Patrick’s explainer: neuroai.science/p/cell-types-e… Dwarkesh episode: dwarkesh.com/p/adam-marbles… Preliminary mapping proposal: ifp.org/mapping-the-br…

How do cell types relate to function? Prodded by @AdamMarblestone's recent appearance on @dwarkeshpodcast, I break down the logic of Steve Byrnes' theory of the steering vs. learning subsystem, and answer why many cell types are better than few for instincts and primary rewards.

Today we launched @CallosumAI. We are building the infrastructure where heterogeneous chips & intelligence co-evolve to solve the world's hardest problems. Today we present our first results. Across four large problem spaces, we break SOTA and deliver orders-of-magnitude improvements in capabilities, cost and speed: 12× cheaper deep context. New web SOTA with open-source, 3x cheaper and faster. 2.4× cache speedups. 1,767× faster tool calling. This is the worst our infrastructure will ever be. We do it by co-evolving heterogeneous chips and multi-agent intelligence - workflows aware of their hardware, models aware of their task graph, kernels aware of their output constraints. An Intelligent System. callosum.com/blog/welcome-h…

New essay: I overanalyze Mac's Night Shift mode and other blue light software filters to find out a recipe for light intake that—in theory—shouldn't wreck your sleep. Colorimeters! Opsins! Corn bulbs! gamma functions! It's all here.