2. Next-token prediction meets video diffusion ▶️

"Diffusion Forcing" method combines the strengths of next-token prediction & video diffusion, training neural networks to handle corrupted data while predicting the next steps. This flexible, reliable sequence model helps produce… pic.twitter.com/QfsYAq7pCQ

— MIT CSAIL (@MIT_CSAIL) 4 mars 2025

2. Next-token prediction meets video diffusion "Diffusion Forcing" method combines the strengths of next-token prediction & video diffusion, training neural networks to handle corrupted data while predicting the next steps. This flexible, reliable sequence model helps produce

→ View original post on X — @mit_csail