AI Updates on 2025-09-13

Gemini app reaches #1 position in the App Store, marking a significant milestone for Google's AI assistant @demishassabis

Google AI Studio sets ambitious goal to enable builders to create 1 million AI-powered apps per day by the end of 2025 @OfficialLoganK
xAI announces major expansion of their Specialist AI tutor team by 10x, hiring across domains like STEM, finance, medicine, and safety @xai
xAI shifts focus from generalist AI tutors to specialist AI tutors, citing significant value addition from the specialized approach @TechCrunch
California passes landmark AI safety bill setting new transparency requirements for large AI companies @TechCrunch

OpenAI announces collaboration with US Center for AI Standards & Innovation and UK AI Security Institute for joint red-teaming and end-to-end testing to improve AI security @OpenAINewsroom

Ethan Mollick demonstrates Claude's ability to create complex PowerPoint presentations from a single vague prompt, including a McKinsey-style SWOT analysis for Hamlet's situation @emollick
Anthropic releases updates to Claude Code SDK with code references, custom tools, and hooks support for faster agent development @_catwu
Tesla AI expands Bay Area ride-hailing service hours, now running until 2am @Tesla_AI

Ethan Mollick discusses the "jagged" nature of AI capabilities, noting that while AI shows graduate-level performance in narrow areas, it remains inconsistent and fails at simple tasks @emollick
François Chollet emphasizes that taste and problem identification skills are more important for researchers than technical ability, cultivated through curiosity and broad reading @fchollet
Qwen3-Next 80B achieves strong performance with only 3B active parameters, demonstrating efficiency in model architecture @Alibaba_Qwen
PyTorch 2.8 adds native XCCL support for Intel GPUs, achieving 99% scaling efficiency on Argonne Aurora and powering Llama3 pre-training at scale @PyTorch
Jim Fan highlights the need for unified robotics benchmarking standards, noting that unlike computer vision and NLP, robotics lacks agreed-upon evaluation protocols @DrJimFan