🏢 ByteDance Seed

This paper enhances Reinforcement Learning from Human Feedback (RLHF) by tackling reward hacking and response diversity issues through improved data construction methods.

Synthetic data can enhance the physical realism of video synthesis, paving the way for more believable generated content.

Expert Race: A flexible routing strategy for scaling diffusion transformer with mixture of experts.

Frac-Connections: An efficient alternative to Hyper-Connections that divides hidden states into fractions.

FlowTok: Seamlessly flows across text and image tokens!

VideoWorld shows AI can learn complex reasoning and planning skills from unlabeled videos alone, achieving professional-level performance in Go and robotics.