Xiangnan He

AgentDoG 1.5 proves you can achieve GPT-5.4-level agent safety with open-source models trained on just 1k samples, slashing deployment overhead by two orders of magnitude.

LLMs for recommendation can now surpass the limitations of static training signals, achieving sustained improvements in ranking accuracy, fairness, and diversity through a dynamically updated Bayesian distillation target.

Automating the translation of economic intuitions into executable computational experiments is now possible, potentially accelerating the pace of economic research.

LLMs can generate recommendations up to 3.1x faster by explicitly modeling token position within items and speculation depth during speculative decoding.

Hallucination mitigation in LVLMs doesn't have to come at the cost of general performance: MPD reduces hallucinations by 23.4% while *improving* overall generative capabilities.

Agent-as-a-Judge can outperform LLM-as-a-Judge in complex environments, but still struggles to reliably verify agent behavior, revealing a critical gap in current LLM-based agent evaluation.

Item agents that self-promote can simultaneously boost recommendation accuracy and fairness, overturning the assumption that these goals are inherently at odds.

Multimodal LLMs get a serious reasoning boost from Durian, a difficulty-aware normalization that tames the instability caused by extreme samples and noisy rewards.

Generative recommendation gets a confidence boost: UGR leverages uncertainty to stabilize training, improve performance, and unlock risk-aware applications.