Current trend: 7b models with the cognitive power of a 70b model. I want to see the explotion of 70b models working as 700b