Meet MoST, a fully open-source AI that understands speech and text in one model. Instead of treating audio and words the same, MoST uses a Modality-Aware Mixture of Experts (MAMoE) to send each token to the right specialists. * Modality-specific experts learn the unique patterns of audio and text. * Shared experts
Chinese AI that can both see and read needs better data. Meet DanQing—an up‑to‑date Chinese vision‑language dataset with 100M image–text pairs. Built from 2024–2025 Common Crawl data and curated with a rigorous pipeline, DanQing aims to power stronger multimodal models. Why it matters * Bridges
AI
AI agents get their superpowers from "skills" - plug-ins that add instructions and code. A new large-scale study of 31,132 skills from two marketplaces finds a big security gap. * 26.1% of skills had at least one vulnerability across 14 patterns in four buckets: prompt injection, data
Models that seem fair in the lab can slip in the wild. Contextual StereoSet shows how measured bias swings when you change the framing—no adversarial prompting required. * Same stereotypes, new frames: Hold content constant, vary time, place, or audience. * Striking shifts: Anchoring to 1990 (vs. 2030) raised stereotype choices
We often map what we can see—buildings, roads, rivers. But what about socially defined places like schools or parks? A new study introduces SocioSeg, a dataset that pairs satellite images with digital maps and pixel-level labels for social categories, organized hierarchically. Built on top, SocioReasoner is a vision-language reasoning
Are smarter AI models safer? A new safety report puts seven frontier models—GPT-5.2, Gemini 3 Pro, Qwen3‑VL, Doubao 1.8, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5—through a unified, cross‑modal test spanning language, vision‑language, and image generation. * Headline: Safety performance
AI
Jailbreak prompts can still push chatbots to produce harmful content, even after safety training. Many defenses either miss sophisticated attacks or over-block harmless requests. This paper finds a useful clue: during generation, models carry latent safety signals that spike when content turns risky, but those signals get overridden by the
When AI agents team up, surprising things happen Large language models don’t start from scratch. They arrive preloaded with vast knowledge and social priors, and they adapt quickly from context. When many such agents interact, their group behavior can surprise us—for better or worse. A new paper argues
AI
AI that reads privacy policies so you don't have to Privacy policies are long, legal, and hard to compare. LADFA is a new AI framework that turns those walls of text into a clear map of how companies collect, use, store, and share your personal data. How it
New paper by Philip Resnik asks a simple question: are today’s language models (like ChatGPT) actually models of how the human mind works? * Implementation level: No. Brains and LMs are built and operate very differently. * Algorithmic/representational level: Weak support. We lack good evidence that LMs use the same
LLMs turbocharge ontology development Ontological Knowledge Bases (OKBs) organize domain know-how, but hand-crafting them is slow and brittle. This research shows how Large Language Models can help experts build and refine OKBs faster—and with more consistency. The team proposes a structured, iterative workflow that uses LLMs to: * extract and
AI
AI agents can be powerful—and unpredictable. AgentGuardian is a new security framework that keeps them on-task by learning and enforcing access rules tailored to each agent. How it works: * Staging phase: It watches the agent run in a controlled setting, logging execution traces and typical inputs to learn what