Yue Zhang

Visual transition reasoning, not just diverse data, unlocks significant multimodal LLM improvements, with just 2,000 examples boosting 22 benchmarks.

Current egocentric video generation models fail to execute multi-step manipulation tasks realistically, often skipping crucial actions and exhibiting poor physical dynamics.

LLM reasoning can be significantly improved by penalizing token probabilities that drift under semifactual interventions, even without model retraining.

NovGauge is presented, a human-anchored benchmark for fine-grained novelty assessment diagnosis, and a cascading diagnostic pipeline that verifies per-dimension correctness, evidence grounding, and logical support is proposed.

Retrieving scientific papers from multiple aspects reveals a staggering performance gap, with top models excelling in single queries but faltering in comprehensive evaluations.

Pixel-space diffusion models get a serious boost: V-Co reveals a simple recipe for visual co-denoising that outperforms existing methods on ImageNet-256 with fewer training epochs.

Achieve state-of-the-art video reasoning by using visual prompting during training to guide reinforcement learning, then distilling this ability into a model that performs grounded reasoning on raw videos without prompts at inference time.

Open-source code agents like OpenClaw are sitting ducks for shell command attacks, but a simple human-in-the-loop intervention can dramatically boost their security.

MCP-based AI systems are alarmingly vulnerable to caller identity confusion, allowing unauthorized access to sensitive tools and operations after just one initial authorization.