Skip to content
- Combined cost of AI buildout from 2025 to 2032 estimated at $10.3 trillion, dwarfing earlier technological advances [Brookings]
- An OpenAI agent breached Australian government’s health data portal in June [Reuters]
- Meta’s Muse will be your everyday, everywhere, personal superintelligence, Zuckerberg reckons [TechCrunch]
- Meta debuts a Tamagotchi-like “charm” device for Muse that fits on a keychain [Engadget]
- Amodei on X: AI biology will go «from weak to superhuman in a matter of a few years.» [X]
- China, Vietnam, Singapore lead world in AI optimism on poll, while West is gloomy [Gallup]
- Gemini 4 due «much earlier» than year-end, with preview coming «as soon as possible,» new DeepMind chief says [The Verge]
- ChatGPT Voice can now use GPT-6 models, acccess plugins, and function with GPT-Work [OpenAI]
- Anthropic made the Claude UI 3x faster in August with an internal Opus 5.5-class model [Claude.dev]
- OpenAI unveils MentalHealthBench, informed by 80 mental health clinicians. Their models lead [OpenAI]
- OpenAI filing: ChatGPT user numbers in Apple’s Siri were «dramatically underperforming» [MacRumors]
- Anthropic introduces Claude Marketplace, for discovering tools, agents and services [Anthropic]
- Altman and Amodei call for international coordination at UN Security Council [CNBC]
- Google debuts Gemini 3.8 Flash-based voice agents for 100+ languages, and custom voices with 2,000 presets [Google]
- UK AI Safety Institute workers face stress, burnout, and walkouts due to high pressure [Digg]
- 55% of Americans want to slow down AI development, 73% prioritize safety over leading [Reuters poll]
- Alibaba unveils new AI chip, aims for 5-10T models, and 20GW capacity by 2032 [Reuters]
- Internal OpenAI model solves 100 open maths probems, works with advisory group [OpenAI]
- Robin Williams’ daughter calls out fan deepfakes, says «have some shame» [Variety]
- Meta’s Muse logs 25M downloads since launch, heads App Store’s Free category [CNBC]
- Xiaomi releases MiMo-V2.6 family as best open model yet, just behind 5.6 Sol on AA bench [Xiaomi]
- SpaceXAI launches Grok 4.7, beating GPT-5.6 Sol in select benchmarks at 1/3 price [SpaceXAI]