iiinova — where your digital self keeps evolving

Patrick Mineault @patrickmineault.bsky.social · 04/05/2026

They move the excitability of cells up and down, both pre-and-post-synaptically. Is cortex like the stomach ganglia made famous by Eve Marder's work, infinitely reconfigurable through modulators and peptides? Scary stuff! 3/3 www.cell.com/neuron/fullt...

090

Patrick Mineault @patrickmineault.bsky.social · 04/05/2026

Do you think that the canonical cortical circuit is complicated? It's so much worse than that! Overlayed on top of the wired connectome, there are *18* different signal-carrying peptides in cortex! Their range is ~similar to the size of a cortical column elifesciences.org/articles/47889 1/

34913

Patrick Mineault @patrickmineault.bsky.social · 25/03/2026

Paradoxically, when you see a yellow on blue slide, you know you're about to get a fantastic lecture. Bonus points for Times New Roman.

1331

Patrick Mineault @patrickmineault.bsky.social · 16/12/2025

✈️ Montreal AI and Neuroscience is over, off to Costa Rica. Great catching up all things NeuroAI with hometown friends. Thanks for the shirt and the bagels, folks!

0140

Patrick Mineault @patrickmineault.bsky.social · 29/11/2025

This equivocation of oscillations and continual learning is a bit of stretch... I'm surprised they didn't get dinged by the reviewers for neuro-babble. openreview.net/forum?id=nbM...

3121

Patrick Mineault @patrickmineault.bsky.social · 06/11/2025

I had way too much fun making these diagrams. Sure, it's not as legible as if I had typed out the text; but I feel like these have a bit more personality

190

Patrick Mineault @patrickmineault.bsky.social · 24/04/2025

Excellent talk from Zico Kolter at ICLR on AI safety. Science of deep learning and AI safety are some of the most impactful ways academic AI researchers can bend the curve

030

Patrick Mineault @patrickmineault.bsky.social · 18/04/2025

I asked o3 for help with LaTeX quote boxes. This is the example quote it used. The machines are on to us.

0603

Patrick Mineault @patrickmineault.bsky.social · 18/04/2025

Thanks, we re-use some of this data in the third meta-plot above. Interestingly, scaling curves match qualitatively and relatively well across different groups. See section 1 where we derive a functional form for the scaling laws that work across sims and Lurz et al. arxiv.org/abs/2411.18526

110

Patrick Mineault @patrickmineault.bsky.social · 17/04/2025

Crowdsourcing this: what are the best examples of scaling laws in foundation models for neuroscience? My top 3 examples: Ctrl+Labs, Sato et al., Mineault et al. @colehurwitz.bsky.social , @mehdiazabou.bsky.social , @nandahkrishna.bsky.social , @averyryoo.bsky.social, @tyrellturing.bsky.social ?

6225

Patrick Mineault @patrickmineault.bsky.social · 12/04/2025

Scott Linderman and I ended up on the same panel at #ODIN2025 at the @alleninstitute.bsky.social , wearing the same J Crew shirt. Twins!

2240

Patrick Mineault @patrickmineault.bsky.social · 04/04/2025

Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems. This paper looks wild: 264 pages and 1416 references. arxiv.org/abs/2504.01990

0181

Patrick Mineault @patrickmineault.bsky.social · 28/03/2025

I am fully aligned with Eva and Blake's vision. I published this piece a couple of weeks ago: www.neuroai.science/p/what-are-f...

A call to action
All of that is a long-winded way of saying that our ambitions in neuroscience–solving all neurological disease, understanding intelligence and consciousness, etc.–are mismatched with our tools and datasets. Foundation models are one area of opportunity: leverage existing and future datasets to find good representations of neural data, make predictions, and optimize in a closed loop, leveraging the differentiability of deep learning models.

That cannot occur in a vacuum, however: the datasets that we collect, and the ecosystem of tools to read and write neural activity, cell identity, synapses, and connections should be matched in a virtuous circle. That will, almost assuredly, involve doing large-scale, hypothesis-free neuroscience focused on tooling and data, including in innovative structures like focused research organizations and coordinated research programs. Hypothesis-free science sometimes gets a bad reputation, but I think it can be most easily justified by epistemic humility: with billions of neurons, trillions of connections, thousands of cell types, hundreds of areas and receptors, maybe the first thing we should do is catalog these dang things, and then figure out how to poke at them to obtain causal models.

I can foresee, 10 years from now, many more neural atlases and databases, distilled into multiple application-specific foundation models and biophysically detailed models, cross-linked and validated through new neurotechnology. At the Amaranth Foundation, we philanthropically fund ambitious neuroscience projects led by ambitious mission-driven individuals; and my role is to lead our efforts in NeuroAI. Of course, I’m not a neutral party in all of this, but I think it’s a compelling vision for neuroscience.

1231

Patrick Mineault @patrickmineault.bsky.social · 27/03/2025

I like to use a diagram on the second slide of my presentations to help people make sense of my unusual background. Unfortunately, I ran out of space with the old one, and 5 circle Venn diagrams are unwieldy. How about this one?

290

Patrick Mineault @patrickmineault.bsky.social · 12/03/2025

Easy hack to have more light in your place: a corn light (100W LED array, equivalent to 1000W incandescent bulb), inside of a photographic light box. It gives about as much light as what I can get from natural sunlight in my house on a sunny day. Will set you back about 60$.

150

Patrick Mineault @patrickmineault.bsky.social · 17/02/2025

Sometimes it's hard to keep track of all the balls you have up in the air. I made a little script to scrape my daily browser history and email subject lines to get a daily digest. Spaced repetition!

2100

Patrick Mineault @patrickmineault.bsky.social · 31/01/2025

Still hasn't cracked pelican on bicycle svg

290

Patrick Mineault @patrickmineault.bsky.social · 31/01/2025

o3-mini is here. I guess they got pulled up the release after DeepSeek took over a news cycle

150

Patrick Mineault @patrickmineault.bsky.social · 27/01/2025

I know I write (overly) long blog posts, but if you scroll down enough you get gems like this

When a fly's compound eye spots a mate in the sky, that's /amore/

0130

Patrick Mineault @patrickmineault.bsky.social · 23/01/2025

I can't wait to see real AI reasoning agents with search, in the line of o1 and DeepSeek-R1. It seems like a step toward building what @michaelnielsen.bsky.social described in his vision of metascience. We might be only a few months away from that scienceplusplus.org/metascience/...

220

Patrick Mineault @patrickmineault.bsky.social · 23/01/2025

Not sure the exact form it would take, but I would like to create some hybrid of a dashboard, an awesome guide or a map of the field e.g. github.com/analyticalmo...

120

Patrick Mineault @patrickmineault.bsky.social · 14/01/2025

Halfway through reading How Not to Study a Disease, about how the amyloid hypothesis has dominated Alzheimer's disease research, this interview from @eboyden3.bsky.social hits hard.

1252

Patrick Mineault @patrickmineault.bsky.social · 13/01/2025

The secret to happiness is low expectations 😎 Fun paper on reward prediction errors and subjective well-being www.pnas.org/doi/10.1073/...

1113

Patrick Mineault @patrickmineault.bsky.social · 12/12/2024

Vision brain scores have not moved much over the last 6 years. Many models with different loss functions can account for about the same variance (e.g. German-to-English vs predictive masked language modeling for language). 5/

120

Patrick Mineault @patrickmineault.bsky.social · 11/12/2024

Anthropic co-founder Jack Clark gave a nice shoutout to our roadmap for NeuroAI safety in his newsletter. open.substack.com/pub/importai...

1162

Patrick Mineault @patrickmineault.bsky.social · 08/12/2024

What's that? Two posts ready to ship on neuroai.science? Is this some kind of a December miracle?

050

Patrick Mineault @patrickmineault.bsky.social · 02/12/2024

The proposals we evaluate span the whole gamut of NeuroAI: biophysically detailed models, embodied and sensory digital twins, cognitive architectures, fine-tuning conventional AI models with brain data, reverse-engineering the loss functions of the brain and mech interp. 5/N

The 7 proposed ways in which neuroscience might help AI safety: Reverse-engineer representations of sensory systems, embodied digital twins, biophysically detailed models, develop better cognitive architectures, use brain data to finetune AI systems, infer the loss functions of the brain, leverage neuroscience-inspired methods for mech interp

1110

Patrick Mineault @patrickmineault.bsky.social · 02/12/2024

We adopt @deepmind’s framework and evaluate 7 concrete proposals to enhance safety without dramatically increasing capabilities. For each approach, we identify specific opportunities for how neuroscientists can contribute to building safer AI systems. 4/N

Three levels of safety where neuroscience might help: robustness, specification and assurance

180

Patrick Mineault @patrickmineault.bsky.social · 02/12/2024

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

Different paths toward safe AI at different Marr's levels

514851

Patrick Mineault @patrickmineault.bsky.social · 18/11/2024

Didn't seem like it used a tool, the result seemed instantaneous:

110

Patrick Mineault @patrickmineault.bsky.social · 28/10/2024

To be clear, I don't think Paul is in the bottom left corner–the panel was quite a bit more nuanced. Paul is very pro naturalistic behavior, and I argued that you need special tooling to analyze that, including foundation models. Hopefully the recording and transcript will be made available soon.

110

Patrick Mineault @patrickmineault.bsky.social · 11/10/2024

I'm glad you posted this, because I needed to tell someone about a celebrity sighting and I couldn't find anybody who cares about classic hip hop. DMC from Run DMC was playing basketball with his crew about two blocks from my place in NYC

110

Patrick Mineault @patrickmineault.bsky.social · 04/10/2024

I know everybody is talking about NotebookLM's podcast feature, but being able to ask questions from dozens of pdfs is the more useful everyday feature. Here it is helping me make sense of the RL/dopamine literature

040

Patrick Mineault @patrickmineault.bsky.social · 17/09/2024

These were the 7 10-year goals laid out by the BRAIN Initiative in 2014. I wonder what the goals will be for the next 10 years. WDYT? braininitiative.nih.gov/vision/nih-b...

041

Patrick Mineault @patrickmineault.bsky.social · 12/09/2024

Had fun over the weekend using a consumer-grade colorimeter (spyderX) to calculate what happens when you turn on Night Shift. I used Claude to translate Matlab code to Python to communicate with the USB colorimeter and measure a LUT. You can download the library here github.com/patrickminea...

010

Patrick Mineault @patrickmineault.bsky.social · 11/09/2024

Where do rewards come from? All rewards are internal. Interesting viewpoint from Barto all.cs.umass.edu/pubs/2009/si...

160

Patrick Mineault @patrickmineault.bsky.social · 08/09/2024

A bistable Gestalt grouping stimulus from Webster and Kay (2012).

030

Patrick Mineault @patrickmineault.bsky.social · 10/12/2023

I promised myself I would eject off of Twitter once I have found a new home, and I'm getting close to there. Thanks for everyone subscribing to the NeuroAI archive ❤️ neuroai.science

071

Patrick Mineault @patrickmineault.bsky.social · 19/10/2023

I've had this e-bike for exactly one month. Since then I've spent more than 15 hours on it and travelled more than 300k (200 miles), mostly from commuting. Bikes are fun!

1101

Patrick Mineault @patrickmineault.bsky.social · 07/10/2023

Trilobites had eyes with with lenses made out of calcite. They are exceptionally well-preserved in the fossil record. Look at these eyeshades! www.science.org/doi/10.1126/...

020

Patrick Mineault @patrickmineault.bsky.social · 30/09/2023

When I tell people I get 9+ hours of sleep, I usually get some raised eyebrows. It doesn't even put me in the top 20% of ppl who need the most sleep! The grindset mindset has lulled us into thinking 5 hours of sleep is normal. It's not! As neuroscientists, we should normalize healthy sleep

050