We are changing how AI scales by changing how it learns.

Backpropagation, the standard learning rule in AI, is upstream of many of the technology's limits: training cost, data requirements, limited long-horizon capabilities, challenges in the physical world, centralized deployment, and the inability of systems to learn continually or improve through use. Computing hardware is also trapped in a local optimum,  around GPUs and their variants, as requirements for memory, bandwidth, precision, power, architecture, and interconnects are all tied to the requirements of backprop and its counterparts at inference time.

We are replacing it.

We have been scaling the primitives of local, plasticity-based, distributed, online learning.

Our team has been leading biologically plausible learning across neuroscience, machine learning, AI accelerators, and memory nanotechnology. We have previously founded unicorns, deployed deep technology in Antarctica and in space, built products used by millions, and sold to dozens of Fortune 500 companies.

We have been building in private. Follow us.