AI neurons run on simple math. Real neurons are far weirder. A new open dataset fitted 1,000+ PyTorch-ready neuron models to real mouse and human cells. Does brain-like diversity make AI better? It's still an open question, but now it's testable. bit.ly/3WfGAwR
Illustration showing a connection between neuroscience and machine learning, featuring diagrams of neurons and circuits. Text reads: "A new bridge between neuroscience and machine learning. Artificial neurons run on simple math, but real brain cells are far richer. Can brain-like diversity make AI better?

ALT Illustration showing a connection between neuroscience and machine learning, featuring diagrams of neurons and circuits. Text reads: "A new bridge between neuroscience and machine learning. Artificial neurons run on simple math, but real brain cells are far richer. Can brain-like diversity make AI better?

Oct 16, 2026 · 11:14 PM UTC

10
53
244
9,617
Sort replies: Relevant Recent Liked
Replying to @AllenInstitute
Without a doubt, I can say from training AI on actual cells it’s the final frontier.
1
7
182
Replying to @AllenInstitute
pytorch-ready neuron models is such a cool idea. curious if anyone tries plugging these into small nets first
1
50
Replying to @AllenInstitute
Fitting over a thousand PyTorch-ready models to real mouse and human neurons turns "brain-like diversity helps AI" from a slogan into something you can measure. Which test would you run first to decide whether that diversity actually helps: sample efficiency, out-of-distribution robustness, or match to real neural activity?
1
139
Replying to @AllenInstitute
oh this is fun. dropping a few hundred of these into a small rnn to see if the heterogeneity helps at all is a weekend project now
1
176
Replying to @AllenInstitute
real neurons being messy might be the point
164
Replying to @AllenInstitute
@Grok find all of the posts where I talk about advances in AI requiring emulation of more types of human brain neurones and even alien neurones which aren't in the human brain. Some of these posts are about scalability not being enough on its own, but mostly they're about heterogeneity of functional artificial neuron type. Some of the posts mention place neurons, mirror neurons etc. Also find the articles in brucerlong.xyz/blog if you are able. There are also some of these posts in my account @BruceRLong
1
1
1
113
Replying to @AllenInstitute
Well said!
19
Load more