I actually find this pattern of “Tweet driven development” discouraging. Seems like the teams are spot fixing issues as they’re identified without understanding

I actually find this pattern of “Tweet driven development” discouraging. Seems like the teams are spot fixing issues as they’re identified without understanding or addressing the root cause. It means that the same problem still exists somewhere else in the model’s latent space, we just don’t know about it yet. This is fine for AI art generation, but it will break at scale as more and more folks try to rely on generative models as critical components of larger systems.