Zhenting Wang

Hallucinated captions can paradoxically boost accuracy in vision language tasks, challenging the notion that they are purely detrimental.

SLIM-RL achieves state-of-the-art performance on math and code tasks with nearly half the training samples required by traditional trajectory-aware methods.

LLM agents can now autonomously generate complex skills with multi-file dependencies, rivaling human-authored skills, thanks to a co-evolutionary verification process that doesn't need ground truth labels.

LLMs can now remember the past without forgetting the details: an indexed memory system lets agents selectively retrieve full-fidelity interactions from an external database, outperforming lossy summarization methods on long-horizon tasks.