Zihan Wang
8 papers ยท Latest:
The JWST early galaxy crisis resolved by a reionization degeneracy
This paper resolves the JWST early galaxy crisis by showing a reionization degeneracy allows standard ฮCDM to explain bright z>10 galaxies.
Jailbroken Frontier Models Retain Their Capabilities
Advanced jailbreaks impose minimal capability degradation on frontier models, challenging assumptions about their safety.
Purifying Multimodal Retrieval: Fragment-Level Evidence Selection for RAG
FES-RAG purifies multimodal retrieval by selecting specific fragments, not whole documents, improving MLLM generation and reducing noise.
Factorized Latent Reasoning for LLM-based Recommendation
This paper introduces Factorized Latent Reasoning (FLR), an LLM-based recommendation framework that disentangles user preferences into multiple factors.
GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents
GLM-5V-Turbo is a new foundation model integrating multimodal perception natively for enhanced agent reasoning, planning, and tool use across diverse contexts.
MEG-RAG: Quantifying Multi-modal Evidence Grounding for Evidence Selection in RAG
MEG-RAG introduces a semantic-aware metric and reranking framework to improve multimodal evidence grounding in RAG systems, enhancing generation accuracy.
Black-Box Skill Stealing Attack from Proprietary LLM Agents: An Empirical Study
This paper empirically studies black-box skill stealing from proprietary LLM agents, demonstrating easy extraction and highlighting overlooked copyright risks.
VEFX-Bench: A Holistic Benchmark for Generic Video Editing and Visual Effects
VEFX-Bench introduces a large-scale dataset, a specialized reward model, and a benchmark for evaluating AI-assisted video editing systems.
๐ฌ Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week โ summarized, scored, and delivered to your inbox every Monday.