Xunliang Cai

Current autonomous agents excel at practical problem-solving but often lack true methodological innovation, revealing critical gaps in their development as independent researchers.

Process evaluations reveal hidden failures in LLM reasoning, showing that lucky successes can mask critical deficiencies in agent performance.

CRPO effectively mitigates exposure bias in self-distillation, leading to superior performance in complex reasoning tasks.

Game solvers can teach LLMs how to make better decisions in long-horizon tasks by providing actionable turn-level feedback, leading to superior performance in complex environments.

Even the best search agents struggle to exceed 35% accuracy on a benchmark designed to push the limits of long-horizon reasoning.

Code agents struggle with evolving user requirements, revealing a 38-point gap in performance across leading LLMs when faced with iterative feedback.

Skill0.5 achieves state-of-the-art out-of-distribution generalization in agentic RL by intelligently combining skill internalization and utilization, outperforming methods that rely solely on one or the other.

Long-context LLM rankings dramatically reshuffle when evaluated across a range of context lengths and capabilities, proving that a single headline score is misleading.

GUI agents can learn world knowledge more efficiently by internalizing causal relationships during mid-training, rather than relying on implicit learning through action annotations or reward signals in post-training.

MONA unlocks faster LLM pretraining and superior downstream performance by turbocharging the Muon optimizer with Nesterov-style acceleration, leaving AdamW in the dust.

LLM agents trained with simulated user and tool noise not only become more robust in messy real-world environments, but also surprisingly improve on clean, idealized benchmarks.

Current LLM agents still struggle to infer and leverage user preferences from fragmented, real-world interactions, revealing a substantial gap between their capabilities and the demands of personalized decision-making.

Open-source LongCat-Video-Avatar 1.5 leapfrogs closed-source competitors in audio-driven video generation by prioritizing practical engineering over architectural novelty, delivering commercial-grade quality and speed.

Interactive world models still have a long way to go: a comprehensive benchmark reveals that even state-of-the-art models struggle to consistently perform well across video quality, interaction adherence, and physics compliance.

Stop LLMs from drifting to English when reasoning in other languages: language-adaptive RL can guide them to stay consistent without sacrificing performance.

Forget brittle orchestration layers – LLMs can internalize complex reasoning as a learnable "HeavySkill" that rivals external agentic frameworks.