Continual Learning via 100B Models and Local GPUs

Continual learning and self-improvement via user interaction may only require 100B parameter models and one dedicated GPU per person.