I’m sure either can do “Draw an owl” style diagrams, but what I mean is a model trained to generate images privately for reasons other than showing them to the user, e.g. to improve spatial reasoning answers

→ View original post on X — @goodside