This Week’s AI & Tech Breakthroughs: Virtual Humans, 6G, and More

ByteDance unveils OmniHuman‑1.5, giving virtual humans logical souls

ByteDance’s digital‑human team launched OmniHuman‑1.5, a new framework that lets virtual avatars not only match lip‑sync but also display appropriate facial expressions, gestures, and logical reasoning based on multimodal large language models and Daniel Kahneman’s dual‑system theory.

Apple’s chief robotics researcher jumps to Meta

Apple’s senior robotics researcher Jian Zhang has moved to Meta’s robotics studio. In addition, three Apple large‑language‑model researchers have left for OpenAI and Anthropic, prompting Apple to consider greater reliance on external AI technologies.

Meituan’s LongCat‑Flash optimizes compute like delivery dispatch

Meituan introduced LongCat‑Flash, a 560‑billion‑parameter model that dynamically allocates compute resources, assigning more “expert” tokens to complex tasks while handling simple tokens with minimal effort, and employs ScMoE (Shortcut‑connected MoE) for parallel scheduling.

6G standardization work officially begins

The 3GPP 6G RAN working group held its first meeting in Bangalore, marking the transition from concept to technical development, with 19 proposals from 17 major tech firms focusing on near‑field communication and high‑frequency bands.

Anthropic bans Chinese‑controlled entities from Claude

Anthropic announced that any company with majority Chinese ownership is prohibited from accessing Claude, extending the restriction to other “adversary” nations, citing legal, regulatory, and security concerns.

OpenAI releases whitepaper on staying ahead in the AI era

OpenAI’s “Leadership Guide” outlines five core principles—coordination, activation, expansion, acceleration, and governance—to help enterprises leverage AI for revenue growth, addressing challenges in adoption, talent, and organizational alignment.

Other tech highlights

Huawei’s openPangu Embedded‑1B achieves state‑of‑the‑art performance on edge devices; Tencent open‑sources the Youtu‑agent framework that requires no model training or paid APIs; Huawei showcases its third‑foldable “Harmony” device, emphasizing ecosystem integration.