17/10/2026
https://www.facebook.com/share/p/1NUZW4L4nb/
The artificial neurons in AI run on simple math units: they add up their inputs and apply a basic function. Meanwhile, real brain cells burst, tire out, and respond at wildly different speeds.
Researchers at the Allen Institute and collaborators have released over 1,000 neuron models fitted to real mouse and human cells and built to train in PyTorch. Early hints suggest human neurons may integrate signals over longer time scales, though the authors call this a first look.
Whether brain-like units improve AI is still an open question. Now researchers have a way to test it.