**************
%20(1).svg)
We believe that everyone should have agency over their own resources - including the ability to provision them to compute networks. In order to maintain that agency at unlimited scale, we build systems for peer-to-peer communication, decision making, and agreement over available data.
Machine learning compute devices convert electricity into floating point operations. We believe that this conversion can be standardised, commoditising compute and releasing software strangleholds on the market. We’re building a truly open compiler stack for machine learning, extensible over all sources of compute - with reproducibility as a core feature.
We believe the only way to successfully outsource a task to another party in a trustless system is to verify the correct execution of the task itself, not to test a quality metric or introduce a proxy competition. In order to achieve verification for ML models, we design and build cryptographic, probabilistic, and game theoretic systems for runtime verification.
We believe that there are more efficient ways to use the world’s compute power than the current silos. Alongside creating the technology to connect, execute, verify, and agree; we also research and build novel distributed learning techniques to create new, modular neural network architectures at a never-before-seen scale.
