iiinova — where your digital self keeps evolving

I don't know how much data we have in neuroscience. The connectomes are fundamentally an advance in data scaling, but much of the in vivo functional data is just searching under the same lamp posts at higher resolution. For some bounded questions it may work. We need a positive control though.

001

And there are a lot of sketchy priors assumed in neuroscience

000

Well, maybe? If we are lucky? If we have only sparse observations of the position of stars and interpreted with some sketchy priors, as we saw from pre Copernican science the simplest "model" may be one that is completely wrong and we would have no way of knowing it.

230

A better test for neuroscience may be can AI learn to predict planetary motion from simply watching shadows on the ground, like Eratosthenes did to estimate the earth's circumference, and stars moving in the sky? We have only indirect measurements for the brain based on circumstance.

000

The metaphor seems completely broken. The ratio of [tens of millions of orbits]:[Newton's Laws] seems inverted to [data we have about the brain]:[The Brain's Complexity]. We don't even know whether the brain data we have is relevant at all for what we need to know.

Brad Aimone @jbimaknee.bsky.social · 15/04/2026

AnD tHe BRaiN is JuST liKe aN AnN... JusT... goTTa ... FINd ... tHe ... bACkpRop... Seriously though... Neuromodulators are cooler than anything anyone in NeuroAI gives them credit for. Maybe norepinephrine and the LC's few hundred neurons is actually all you need.

Brad Aimone @jbimaknee.bsky.social · 04/04/2026

We should fund science more not less. But the cynical part of me thinks this is an opportunity to maybe stop funding the same neuroscience questions with just a fancier microscope or just one region over time and time again. We need to cure diseases and fix AI. Not keep doing the same thing.

Brad Aimone @jbimaknee.bsky.social · 23/02/2026

Excited to be in San Antonio for the UTSA AI Matrix THOR Neuromorphic Commons kickoff! THOR will be one of the first community resources fully dedicated for accessing scalable neuromorphic hardware! Check it out! #NeuroAI #Neuromorphic 🧪🧠🤖 www.neuromorphiccommons.com/events/thor_...

040

Brad Aimone @jbimaknee.bsky.social · 09/02/2026

Am I alone in being skeptical when seeing "We need WORLD MODELS because that's what the BRAIN does!"? Won't this just be another 'AI tech bros use the brain to get attention and $$$ but ignore it at the first opportunity'? Why trust any of the AI crowd to talk about the brain? We need real #NeuroAI

000

Brad Aimone @jbimaknee.bsky.social · 09/02/2026

How a place can ruin both coffee and donuts astounds me

000

Brad Aimone @jbimaknee.bsky.social · 07/02/2026

I've always understood the "have to start somewhere" and the *hope* that visual cortex is all we need since it is easy to access and easy and intuitive for inputs. But we're, what, 75 years into V1's reign over neuro? With little generalizable to neural disorders to show for it? Let's move on.

Is there even such a thing as academic machine learning anymore?

111

If you defer leadership to industry, you relinquish any right to criticize the outcome being profit-centric. Kudos to the BRAIN leadership for embracing the brain / AI / computing connection. That takes courage because so many neuros and AI tech bros deny the connection out of self-interest.

000

Maybe cynical, but any time a scientist says "We should leave AI to industry" it is because they are scared about $$ going to something they don't work on I've heard for >10 years "let industry lead" and we have LLMs, power plants & data centers #NeuroAI needs research, not just venture capital

150

Brad Aimone @jbimaknee.bsky.social · 04/01/2026

There is plenty of space to innovate in #NeuroAI. The issue has been that neuroscientists don't even try as they assume industry will do it. Deferring AI and neural computing to an industry that only cares about selling ads is not a way to help further our understanding of the brain.

131

You can take BlueSky out of Twitter, but you can't keep the Twitter tech bros out of BlueSky

000

Brad Aimone @jbimaknee.bsky.social · 04/01/2026

Well, we have another 20 years until the cortex field rediscovers a hippocampus finding.

Brad Aimone @jbimaknee.bsky.social · 03/01/2026

I agree. Deep learning has huge value on its own. But it isnt brain inspired in intent or practice I sometimes see "debates" with LeCun or Dally, what is the point? To convince them? Of what? It's a different field Neuro may be able to overcome ANN limitations, but the brain path won't come from DL

000

It's unfortunate, but there really isn't any other monetization path that justifies the insane capital expenses they're investing in.

010

In the end, I try to be practical about all of this. Philosophically we can debate about what true understanding means, but practically we want better and smarter AI algorithms and to be able to fix the brain. How do we do that? If digital isn't sufficient, what is the scalable alternative?

000

I'm not a Turing worshiper, but I think you're fixated too much on the discrete/analog thing. The brain isn't analog all the way down, synapses and ion channels are really stochastic discrete elements. Really the stochasticity, not the continuity, is where the brain diverges from classical computing

200

You're not guilty of this to begin with. :) That's like me saying my New Year's resolution is not to start every day off with a Bloody Mary.

110

Well, I certainly believe that there are better models to use than serial Turing machines. Though "analog!" is a pretty weak alternative for a number of reasons. But saying Turing computation fundamentally cannot represent what the brain is doing is a very high theoretical bar to get over.

100

The brain clearly solves the same types of problems (control, inference, etc) in a different way. The same functions but different algorithms on a different model of computation. It isn't marginalizing the brain to say that it computes, it helps demystifies it. Which is what we have to do.

000

This is the danger of falling into the trap of implementations. Today we use a certain type of computer to do scientific computing and AI; but that doesn't mean that Von Neumann machines are the only type of computer or that sequential linear algebra is the only type of math that is useful.

100

Assuming the brain is representing the world for decision making and survival, it is effectively modeling the world with neurons. That's exactly what numerical computing is - modeling something with something else - the substrate is just different than transistors in a stored program architecture.

100

The brain isn't doing Runge-Kutta in floating point on a von Neumann architecture, but that doesn't mean the principles of applied math and theoretical computer science don't apply. The brain isn't magic. Math and computer science apply to it, just like the laws of physics do.

200

Basically, I claim that neural computation is just any other numerical method, with limitations like any other and amenable to analysis like any other. We simply don't yet know what that method is.

000

I get what youre saying but the brain is representing the world (external and body) inexactly. Whether digital or analog, discrete or continuous, it doesn't matter. The brain is approximating some other dynamics with its dynamics. That approximation has numerical limitations like anything else.

200

In fact, Id go so far as to argue that many neurological disorders are a breakdown of that robustness of neural computation that we take for granted.

000

That I disagree with. If the brain is computing, it has to do so reliably. Which means the same numerical stability issues matter. If I see a cat, I should always perceive a cat. And we do. Even if the underlying dynamics are chaotic. To me, that is one of the biggest open questions in neuro.

200

My takeaway, which has stuck with me for ten years since, is that the brain's computations must be abstracted from the microsecond details of biophysics we can potentially measure. The timing of spikes matters, but relatively across a population, not individually.

000

Many years ago we had a study (file cabinet paper sadly) that showed no matter the numerical precision, the stiff nonlinearities and fanout of recurrent spiking circuits made simulations diverge. At 1st this says simulations don't work, but the brain also has to operate reliably with such stiffness

The goal of neuroscience and NeuroAI really shouldn't be to have an exact in silico clone of an individuals brain. That isn't necessary for almost any helpful societal impactful application of neuroscience research

120

This boils down to what the purpose of the simulation is. Weather predictions are fine as samples over days because that saves lives and $$. I'd argue that to help treat disease - finding the locus of seizure generation, say? - a sample over short time is fine. Perhaps also for language generation

100

We don't *know* what the right abstractions are for the brain. That is a huge problem. And we likely don't have the data to constrain even properly abstracted models. That is a huge problem. But none of that has to do with digital computers being wrong for the brain.

100

Again, every field of science and engineering does this effectively aside from neuroscience. They don't simulate every air atom at quantum resolution in aerodynamics simulations of Boeing planes. Abstractions are powerful and often okay.

100

Reasonable people can disagree about what level of fidelity is required for any simulation. I fully agree that the brain's complexity and near-chaotic dynamics make extremely precise simulation extremely challenging. But we likely don't have to simulate at that level of detail.

100

For sure-that's critical. But 99% of PyTorch users couldn't derive backprop on a white board, much less program from scratch. So the entire field is based on codified primitives that are increasingly abstracted away from how the brain works. And that's fine for today's AI, but likely not tomorrow's.

Our problem with simulating the brain is 99% not having enough data, and maybe 1% the fit of computer hardware; modeling the brain is like trying to predict the trajectory of a hurricane based on a couple boat weather reports This is rapidly changing though with neural data being collected today

There is nothing fundamental about the brain's dynamics that make it incompatible to computers. The entire world is continuous real-time dynamics, yet we simulate every other type of physics on discrete binary computers, from weather forecasting to protein folding. The brain is no different.

Yeah, they really don't want to think about it. Partially it is fair, because they have good models without spikes. But they're stupid expensive (real $$$$), and they have to know that they can't keep going like this forever. But they're really scared of biology.

Optimistically I do think things are changing. At NICE every year we have seen a maturation of thinking about both hardware and algorithms that portends a revolution of thought. Truly brain-inspired spiking algorithms are the future of AI, but they require more than just the PyTorch way of thinking.

100

Ultimately, who will care? Everyone *not* at NeurIPS. Studying how LLMs process information isn't going to help us understand how the brain works. Fixing them won't give us cluse for curing diseases. But people need cures. Making AI efficient won't help Nvidia's stock price, but people need that. 5/

160

It is time to stop letting the AI community define our field. They don't care about the brain, they don't want energy efficiency, and they don't want smaller models. They all stand to lose too much $$$ to that type of competition. Seriously, stop trying to impress people who want us to fail 4/

280

The field is starting to see in spiking state space models that populations of neurons can encode complex memories spatiotemporally. We have seen in our NeuroFEM linear solver that the full embrace of spiking enables efficient numerical computing without loss of accuracy. And so on... 3/

120