When GPT-3 meta-learns, the weights of the design do not change, but as the design computes layer by layer, the inner figures become new abstractions which can carry out responsibilities it has hardly ever completed right before in a perception, the GPT-3 model with the 175b parameters is not the genuine design-the serious model is those people ephemeral numbers which exist in involving the enter and the output, and define a new GPT-3 tailored.