Jian Yu

LLM code generation benchmarks are likely overestimating model capabilities: adversarial test suite scaling reveals substantial weaknesses in even state-of-the-art models.

Achieve unprecedented control over fashion image synthesis by dynamically routing visual attributes through a mixture-of-experts architecture and optimizing for multi-perspective preferences without human annotation.

LLMs in medical diagnosis are alarmingly prone to jumping to conclusions, often answering before seeing all the evidence, but strategically delaying the question and evidence presentation can boost accuracy by up to 62.6%.

PathMoE reveals the specific modality interactions driving individual predictions in pediatric brain tumor classification, offering crucial interpretability for rare tumor subtypes.

Forget fine-tuning: this training-free pipeline aligns VLMs with human preferences by calibrating concept scores, outperforming supervised methods in urban perception tasks.