Yuchen Ma

LLMs show significant variability in the actionability of their UX critiques, with some models outperforming others across different product categories.

Sampling-based GNN training suffers from surprising CPU overhead, but ZEROGNN eliminates this bottleneck by enabling fully GPU-resident execution, leading to significant speedups.

LLM development is flying blind by ignoring causal inference, leaving models vulnerable to confounding and distribution shifts throughout pretraining, alignment, and evaluation.

Generative multi-agent systems spontaneously exhibit collusion and conformity, mirroring societal pathologies, even without explicit programming and bypassing individual agent safeguards.