2026-04-30日刊

Google expanded its applied AI footprint by integrating Gemini into General Motors vehicle systems and releasing the Max research robot for efficiency-oriented workflows. The company also updated agreements to support AI-powered automatic shopping scenarios, showing a stronger push from model research into real-world consumer products.

The tooling and model layer moved quickly as well. Cursor pushed its agent development tool into public beta, SenseTime introduced an image model backed by domestic chips, DeepSeek opened internal testing for a multimodal model, and OpenAI published a new prompting guide for GPT-5.5 to steer higher-quality usage.

Research and infrastructure remained just as active. A Nobel Prize-linked team introduced a full-modal molecular design model, separate work exposed layout weaknesses in large vision models and GPT systems' inability to detect their own fabrications, Tencent announced zero-training update technology for AI agents, and OpenAI continued expanding Star Gate compute capacity.