Hao Lin

Results show that CoSTAR enables small, locally deployable LLMs to achieve performance competitive with substantially larger LLMs for privacy-sensitive COBOL legacy systems.

Co-training item embeddings in recommendation systems can boost prediction accuracy and engagement while slashing computational costs by 30%.

Explicit semantic augmentation can boost LLM-based code translation success rates by over 200%, especially for smaller models.

LLMs can reason more efficiently by sharing intermediate thoughts during parallel search, achieving better accuracy with less computation.

Omni-modal LLMs can ace captioning and QA, but AVID reveals they're surprisingly bad at spotting audio-visual inconsistencies in videos, a crucial skill for trustworthy AI.

CLIP can now understand "no dog" without any fine-tuning, thanks to a plug-and-play module that disentangles negated semantics and penalizes false positive matches.

Key contribution not extracted.