Topics

LLMs hallucinate confidently. Uncertainty estimation flags errors in NLP & code—strong on open-source models, misses nuances in GPT-4o. A plug-and-play step toward safer, self-aware AI.

Why do LLMs sound smart but often get it wrong? A decades-old knowledge system offers a blueprint for truly dependable AI.

Microsoft Research Asia has introduced VASA-1, a groundbreaking framework for generating lifelike talking faces with appealing visual affective skills (VAS) given a single static image and a speech audio clip.

Hunyuan-Large rewrites the rules for scalable AI—52B parameters, 256K-token context, and a design that rivals GPT-4 without the compute drain.

LLMs can predict and emulate positive human ethical judgment, but also be used as a manipulative tool to twist and distort ethics.

Alibaba's Kimi K3 delivers frontier AI at lower cost than GPT-5.6 and Claude, signaling a new era for Chinese open-source AI and enterprise large language models

What if AI organizes knowledge like the human brain—or even the universe? Researchers uncover stunning geometric patterns shaping how large language models understand the world.