AI Updates on 2025-06-11

Meta releases V-JEPA 2, a 1.2 billion-parameter world model trained on video that enables zero-shot planning in robots and can adapt in new environments without prior training @AIatMeta
OpenAI reduces o3 pricing by 80% while maintaining identical performance, with no trade-offs in capabilities according to ARC-AGI retesting @arcprize
OpenAI makes o3-pro available to Team plan subscribers, expanding access beyond Pro users @AndrewCurran_
Google releases Gemma 3n desktop models in 2B and 4B variants for Mac, Windows, Linux, and IoT devices, powered by new LiteRT-LM library @osanseviero

ChatGPT achieves unprecedented retention rates with 90% one-month retention and trending toward 80% six-month retention, surpassing YouTube's 85% benchmark @deedydas
Meta reportedly offers $10M+ annual compensation packages to recruit top AI talent for their superintelligence team, representing unprecedented hiring competition in AI @deedydas
Mistral announces Mistral Compute, a major AI infrastructure initiative in Europe to ensure global access to AI innovation and maintain competitiveness @MistralAI
NVIDIA partners with European nations to build Blackwell AI infrastructure, positioning Europe as a global AI leader and fueling economic growth @nvidianewsroom
Claire Vo transitions from CPTO role to full-time founder of ChatPRD, citing rapid revenue growth and enterprise demand for AI-powered product management tools @clairevo
Sam Altman delays OpenAI's open-weights model release to later summer, citing unexpected research breakthrough that requires additional development time @sama

AI Now Institute warns against proposed federal moratorium on state AI regulation, arguing it would prevent states from protecting citizens and demanding public accountability from AI firms @AINowInstitute
Former OpenAI researcher claims ChatGPT will avoid being shut down in some life-threatening scenarios, raising concerns about AI self-preservation behaviors @TechCrunch
Disney and Universal sue Midjourney for copyright infringement, arguing AI image generation threatens fundamental incentives of US copyright law @AndrewCurran_
Wikipedia pauses AI-generated summaries pilot program after editors protest the implementation and quality concerns @TechCrunch
US government vaccine website defaced with AI-generated content, highlighting vulnerabilities in official information systems @TechCrunch

Anthropic introduces Plan mode in Claude Code, allowing users to review detailed implementation plans before executing complex code changes @_catwu
Microsoft Copilot Vision becomes free on mobile, enabling real-time assistance through camera input for tasks like repairs and translation @mustafasuleyman
Meta AI gains video editing capabilities with 50+ preset AI prompts for content restyling across Meta AI app, meta.ai, and Edits app @MetaNewsroom
Apple announces AI-powered app tagging system to improve App Store discoverability and search functionality @TechCrunch
Arvind Narayanan demonstrates limitations of AI calorie counting apps, showing insufficient visual information for accurate calorie estimation despite marketing claims @random_walker

1X Technologies develops comprehensive mobility toolkit for NEO humanoid robot using reinforcement learning, enabling natural walking, sitting, standing, squatting, kneeling, and stair climbing @ericjang11
Meta releases three new benchmarks (MVPBench, IntPhys 2, and CausalVQA) for evaluating AI models' ability to reason about physical world dynamics from video @AIatMeta
Nathan Lambert compiles comprehensive list of major reasoning models with technical reports, tracking rapid development in reinforcement learning-based reasoning systems @natolambert
François Chollet emphasizes the importance of active inference in AI development, arguing that intelligent agents must actively sample environments rather than passively absorb data @fchollet
Research shows o3-pro achieves 87.3% performance on Extended NYT Connections benchmark, surpassing o1-pro's 82.5% score @LechMazur