Generative Vision Interview Questions, issue 12, Jun 20, 2026
The DiT Scaling Paradox
Senior AI Engineer interview at OpenAI, and the interviewer asks:
“You doubled your DiT’s parameter count expecting FID to drop. It barely moved. What did you miss?”
Don’t say: “We need more data or longer training.”
The hidden dimension where identical parameter counts yield wildly different compute, and why tracking Gflops instead of weights is the secret to breaking the image quality plateau.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.