Search papers, labs, and topics across Lattice.
Peking University
3
0
8
Stop obsessing over state prediction accuracy in text-based world models: aligning them with *behavior* yields better long-term planning and evaluation.
Unsupervised RL for text generation doesn't have to collapse into gibberish: rewarding relative information gain between specialist and generalist policies unlocks meaningful content creation.
Current AI-generated image detectors crumble when faced with the structured complexity of scientific figures, revealing a blind spot in our ability to maintain research integrity.
