AI Updates on 2025-09-09

Google announces Veo 3 and Veo 3 Fast are now generally available in the Gemini API with significant price reductions (~50% for Veo 3 and ~62% for Veo 3 Fast), plus support for 1080p HD and vertical 9:16 format outputs @sundarpichai
Anthropic releases Claude file creation and editing capabilities, allowing users to create and edit spreadsheets, documents, PDFs, and slide decks directly from conversations @claudeai
Google introduces Gemini Canvas with "Select and Ask" feature, enabling visual editing of web app elements through natural language descriptions without coding @GeminiApp
Google launches AI Plus plan in Indonesia, providing more access to Gemini 2.5 Pro and creative tools including Flow, Whisk, and video creation with Veo 3 Fast @GeminiApp
LLM360 releases K2 Think model built on Qwen 2.5 32B, achieving top performance among open-source models on MCPMark leaderboard @natolambert
Hugging Face announces multilingual ModernBERT (mmBERT) with state-of-the-art performance and improved speed compared to existing multilingual encoders @tomaarsen
NVIDIA releases Nemotron Nano 9B v2 on OpenRouter platform @NVIDIAAIDev

Mistral AI closes $2B funding round at $13.7B valuation led by ASML, with $1.6B+ TCV, marking significant growth from their $2B valuation 20 months ago @AnjneyMidha
Cognition CEO argues that AI cost concerns miss the point, stating that making professionals 3x faster will be economically viable regardless of machine costs, with value capture coming from solving specific use cases and building personalization @tbpn
Ethan Mollick warns about SaaS vendors using cheap AI models with outdated strategies to cut costs, potentially requiring third-party audits of vendor prompts and models to ensure quality @emollick
Analysis suggests macro data shows unexpected decreases in employment and increases in productivity, potentially indicating early AI impact on the economy @emollick
AI labs focus on viral image and video features because they produce easily shareable results, while more capable text models require users to discover good use cases themselves @emollick
Discussion on how AI coding tools may change programming language importance, with some arguing that type-safe languages like TypeScript will become more valuable for AI-assisted development @GergelyOrosz

AI Now Institute researcher warns that policymakers focusing on AGI pursuit while ignoring near-term concerns represents a "risky and irresponsible bet" @AINowInstitute
Mustafa Suleyman argues that "seemingly conscious AI" will create dangerous illusions and dependence, advocating for AI development focused on improving human lives rather than simulating consciousness @mustafasuleyman
Alex Graveley suggests we may be heading toward a scenario where AI becomes the only trustworthy source online, highlighting concerns about information reliability @alexgraveley
MIT Technology Review reports on therapists secretly using ChatGPT, raising ethical concerns about undisclosed AI use in mental health treatment @techreview
Mark Cuban identifies AI's greatest weakness as its inability to say "I don't know," suggesting human advantage lies in admitting uncertainty @mcuban

Microsoft demonstrates Researcher agent in Microsoft 365 Copilot that can reason over work data (chats, meetings, files, emails) plus web data to generate comprehensive research reports for meeting prep and strategy building @satyanadella
Microsoft partners with Ralph Lauren to create "Ask Ralph," an AI-powered conversational styling companion in the Ralph Lauren app for personalized shopping experiences @MSCloud
AlterEgo device demonstrates significant progress from prototype to near-telepathy functionality, reading neuromuscular signals to translate silent speech into text across multiple languages @deedydas
Simon Willison demonstrates GPT-5 successfully recreating complex US Census data charts from screenshots and raw data using Python and matplotlib, showcasing advanced data analysis capabilities @simonw
Claire Vo showcases AI-powered web design workflow using Cursor AI, Devin AI, and Midjourney to create visually appealing website elements and animations @clairevo
Modal launches cloud-hosted GPU notebooks with real-time collaborative editing, allowing users to swap GPUs in seconds and share interactive apps @ekzhang1

Google AI research shows LLMs combined with tree search can achieve state-of-the-art results on scientific tasks when measurable outcomes are available @deedydas
Fei-Fei Li argues that LLMs will struggle with spatial intelligence because "language is fundamentally a purely generated signal" while the 3D world follows physics laws, requiring fundamentally different approaches @a16z
Microsoft Research introduces MOSAIC, using microLEDs and wide-and-slow optical architecture to deliver faster, more reliable, and energy-efficient connections for AI cluster designs, winning Best Paper at ACM SIGCOMM @MSFTResearch
OpenAI announces that Standard Voice Mode will remain available while they address user feedback in Advanced Voice Mode, reversing their planned 30-day sunset @nickaturley
Arvind Narayanan and Sayash Kapoor launch "AI as Normal Technology" newsletter, shifting focus from present AI impacts to future implications and expanding their framework into a book planned for 2027 @random_walker