Today’s hardware used to train artificial intelligence (AI) is not well suited for the task. This becomes particularly clear when you look at the floors of server racks in today’s data centers, which have the energy requirements of a small town and yet cause long runtimes and thus long development cycles. The reason for these inefficiencies is the attempt to reproduce the processes of the human brain by software, but then to execute them on an infrastructure (digital computer), which is fundamentally different in its mode of operation from that of the human brain.
Therefore, GEMESYS is working on the development of a chip, whose mode of operation is much closer to the human brain. They expect an increase in efficiency of up to 20,000x compared to the current state of the art.