Icarus Cybernetics

Origin

I have been coding since I was five, and doing machine learning since middle school, trying to understand one thing: what is the theoretical limit to machine intelligence? That took me into research with Harvard Medical School, Carnegie Mellon, and MIT CSAIL, focused on mechanistic interpretability to understand where these models hit their information-theoretic and thermodynamic limits.

My cofounder Rayne came at it from the other side. A computational biologist set on modelling the human mind algorithmically, a multiple-time olympiad medallist, recognised internationally for his work on Darwinian theory. His research is what Icarus Cybernetics was built on.

We became roommates, and had many ML breakthroughs. They converged on one architecture that did what each of us had been chasing independently. We called it Apex. It clears every bottleneck holding modern AI back.

By then it was clear where everyone else was stuck. New labs split two ways: chasing something radically different but narrow, which turns into a gimmick, or building on the transformer, and inheriting every one of its limits. The only way to beat the best labs is to be better than all of them, at everything, with an approach no one else has. This is our bet.

Once we understood Apex's capabilities, recursive self-improvement, intrinsic motivation, and more, we began developing the cybernetics layer that merges it with the human mind, to reach a level of intelligence equal to the smartest AI that will ever exist.

In the short term, we want to show the world what a radically different, non-backpropagation approach to machine learning can do. In the long run, that layer is how people stay relevant in a world of superintelligence.

In our lifetime, we want to build the foundation for the intelligence it takes to solve the problems most people call impossible, and to move humanity forward on a scale it has never seen.

Prabhdeep Singh
Co-Founder and CEO