At Atombeam we have long appreciated both the incredible capabilities of Large Language Models and the inherent weaknesses of the transformer-based architecture most are based on. Collectively, we have all experienced their limitations as well. When stateless LLMs are used for tasks that require true learning, hallucinations and incorrect answers are inconceivably accepted as a matter of course.

It has also become overtly clear that current state of AI – LLMs that require as much power as a small city in part because they start each query from scratch – is not sustainable, even as the sector earmarks an almost inconceivable amount of money for research, much of it focused on incremental improvements on the status quo. 

The article by Atombeam’s CEO Charles Yeomans, “Why The Need for AI R&D Is Greater Than Ever” in Forbes explores this last point in  detail and what is becoming inescapably clear: incremental improvements in AI will not suffice. What is needed is a new design.

We believe our Persistent Cognitive Machine is this new architecture, one that is fundamentally different from the ground up, and dynamic by design.