We get asked a fair question often. If the idea works, where is the proof. The honest answer has two parts, and we think both are worth stating plainly, because the gap between them is exactly where the interesting work lives.
What the research established
The foundation is not a pitch. It is a body of peer-reviewed work on energy-based networks, the class that includes restricted Boltzmann machines and deep belief networks. Across that work, quantum-assisted training showed three results that hold up under review.
It converged faster than the classical contrastive divergence baseline. It reached better optima, settling at lower free energy than the classical method found. And its sampling cost stayed flat as the network grew, while the classical cost climbed with size. That last property is the one that matters most, because flat-versus-rising is the difference between a trick and a scaling advantage.
This is not folklore. The underlying research carries more than 1,700 academic citations, and the method is protected by a granted US patent, US12566987B2, assigned to Cornell and running to 2043.
The distance we will not paper over
Now the honest part. Energy-based networks are not transformers. A result on restricted Boltzmann machines is a proof of mechanism, not a benchmark on a modern language model. Anyone who tells you the transformer-scale numbers already exist is selling something. They do not exist yet, because building them is hard, and hard is the whole point.
So we say it the way it is. The mechanism is proven in published research. Carrying it to the architectures that labs train today is our current work. We would rather be trusted on a smaller, true claim than admired for a larger one we cannot yet stand behind.
That is the posture the whole company is built on. The science is real and public. The scale-up is genuinely unsolved. The distance between the two is not a weakness to hide. It is the reason there is a company here at all.