Yeah, I get what you're saying but both are challenging the current MatMul methods. The L-Mul paper claims "a power savings of 95%" and that is the thread topic. Bitnet proves that at least 70% is possible by getting rid of MatMul.