Baichuan Intelligent Releases Character Large Model Baichuan-NPC


The Shanghai Internet Association has established an AI Working Committee and launched a pilot project to enhance AI safety supervision, aiming to improve corporate governance. Shanghai Mole Network Technology was appointed as a deputy director unit to support this development.....
On May 8, the MIIT, the Administration for Market Regulation, and the Ministry of Commerce jointly released the national standard 'Grading of Intelligence for Artificial Intelligence Terminals', establishing a unified evaluation system. The standard adopts a '2+N' framework, clarifying the definition of smart terminals, grading logic, and testing methods. It sets up four levels of capabilities from L1 Response Level to L4 Collaboration Level. The L4 level will be further improved with technological development.
Mozilla engineers used Anthropic's Claude Mythos AI model to discover and fix 271 security vulnerabilities in Firefox version 150, including 180 high-risk, 80 medium-risk, and 11 other-level flaws, with browsing potentially affected by high-risk ones.....
A Microsoft report from 2026 shows that 17.8% of the working-age population globally uses generative AI, but the gap between developed and developing countries has widened. In developed countries, 27.5% of people aged 15-64 use the tool, compared to only 15.4% in developing countries, with the gap increasing by 1.5 percentage points since the second half of 2025. The main reasons are differences in internet access, digital skills, and electricity availability.
OpenAI launches three real-time voice models: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper, integrated into the Realtime API. GPT-Realtime-2 is the first smartest AI voice model, addressing high latency, inability to naturally interrupt, and multilingual support challenges in voice interactions.....