Thread by @isaakfreeman

I’m leaving MIT and not continuing into my PhD. AI is coming too fast for humans to keep up.

But there might be a way: I realized digital humans are more possible than most think. With capable AI researchers helping, maybe for $10B, maybe in less than 10 years, on 50k H100s.

Running a human brain might need only ~50,000 H100 GPUs. xAI already has 200,000+ H100s or better.

To anchor the discussion, I did some very rough napkin math: Under fairly pessimistic assumptions using current high-resolution neurons (eg Hodgkin-Huxley), multi-state synapses, a human brain might be in reach of ~600 exaFLOP/s of compute, 700 GB memory storage per GPU, and 24 GB/s interconnect bandwidth. That's already in reach for today's clusters!

If much simpler neuron models (eg Leaky-Integrate-and-Fire) are enough--which needs more research to be empirically determined--then a human brain might be as cheap as ~2-3 petaFLOP/s. That's nearly a single H100 at FP16. (Memory and interconnect are likely tighter constraints.)

But what neurons to run? What parameters? What connectivity?

Hence data generation is the actual bottleneck with many problems to solve: We need hundreds of next-gen microscopes running for years. Automated large-scale tissue collection and staining.

~20x Expansion microscopy with thorough molecular staining of 30+ receptors, neurotransmitters, and neuropeptides. X-ray microscopy to eventually image a whole human brain in less than a year.

Whole-brain functional imaging scopes able to image brain activity across worm and fish brains to crack structure-to-function translation.

Also we'll need structure-to-function prediction models, proofreading models, rigorous benchmarks, and thorough animal emulations as proof-of-concept.

The field is forming: Early attempts at worm emulation like BAAIWorm, a 140,000-neuron fly connectome, incomplete attempts at a fly emulation going viral on X, datasets generated by BCI work, an incoming zebrafish connectome, microscopes able to image at gigahertz speeds...

With the intention of making the field accessible, I used my last time at MIT to sketch out an in-depth thesis of how to go from worms to digital humans. While rough & imperfect, I poured lots of love into it, so I hope this is enjoyable to read!

pdf.isaak.net/thesis

Huge thanks to everyone answering my many questions. Thanks for the relentless writing sessions, mentorship and pivotal convos throughout: @eboyden3 @kesvelt @geochurch @tylercowen @mxschons @nc_znc!

If you’re excited to work on this or fund this: Say hi at [e-mail].

Because in the age of AI capabilities, humans are falling behind. We've been working on something new... more soon!