Zhe Wang

Multi-turn interactions can be effectively optimized in LLMs using tailored RL strategies, overcoming significant challenges in credit assignment and reward design.

Learning the generation order in multimodal tasks can boost performance by over 4%—a game changer for DLMs.

Bridging the gap between synthetic and real-world motion prediction, this framework achieves superior performance by leveraging objectness priors to refine motion labels.

ATCCaps reveals that effective call-sign recognition in ATC communications hinges not just on quantity, but on the quality and accuracy of audio-text supervision.

Pipette transforms wet-lab robotics by boosting training efficiency and accessibility, achieving up to 74.7% success rates with just 30 demonstrations per task.

Learning insertion order in sequence generation can drastically improve modeling quality and generalization, challenging the dominance of fixed-canvas methods.

Frontier models can't build playable games in one shot, but a closed-loop system using GUI agents to playtest and refine code achieves a 66.8% success rate, proving that game generation needs to be a conversation, not a translation.

Pathologists using AI assistance can boost diagnostic accuracy by 8% and slash diagnostic time by 20% in lung pathology, according to a prospective, multi-center RCT.

Current video generation benchmarks miss the forest for the trees: EvalVerse actually measures cinematic quality, not just prompt adherence.

Forget wrestling with finicky text prompts or tedious manual camera paths: ShotVerse lets you generate cinematic multi-shot videos from text, thanks to its clever "Plan-then-Control" framework and a dataset of aligned camera trajectories.