[hamilyon2]

The startup that is able to train new models from the ground up with new architectures of course. Nobody assumes that transformer-self attention is pinnacle of llm design we will not be able to surpass it, right?