Today's Top 30 AI Stories

Mustafa Suleyman predicts human-level AI performance on most professional tasks within 18 months, including accounting, legal, marketing, and project management. He frames this as the path to superintelligence and rapid model creation. The forecast comes amid massive Microsoft AI investments.

A new GitHub repo with 10.7K stars offers 200+ models including Sora, Kling, Veo, and Flux with cinematic controls for free. Users are canceling Runway and Higgsfield subscriptions in favor of the local, no-paywall alternative. It integrates directly with Claude and Codex for end-to-end production.

The ex-OpenAI researcher files massive put options on Nvidia, AMD, Broadcom, and others worth over $7B while holding power and infrastructure longs. He believes semiconductor valuations have peaked even as AI demand grows. His Q1 2026 13F shows a near-tripling of his portfolio.

Anthropic acquires Stainless, the platform behind every Anthropic SDK and MCP server since day one. The move strengthens tooling for developers building reliable AI agents and applications. It underscores Anthropic's focus on production-grade agent infrastructure.

Leaked details show Gemini's new desktop app with Spark mode for local agentic tasks, folder access, code execution, and 'Stream to Cursor' context awareness. Internal references to Veo 4 Omni and custom skills point to deep multimodal integration. The update positions Gemini as a proactive desktop agent.

NVIDIA open-sources a 2.6B parameter world model that turns single images and text into controllable 3D environments on a single GPU. It supports robotics, simulations, and embodied AI research with full code and paper available. The release lowers the barrier for world-model experimentation.

Zhang Yiming, founder of ByteDance, is investing heavily in Unitree Robotics and ByteDance AI Labs. This contrasts with India's richest focusing on consumer products while US and Chinese leaders double down on frontier tech. The moves highlight diverging national AI priorities.

Bryan Johnson outlines how AI would accelerate genome sequencing, AlphaFold protein mapping, drug screening, and decentralized vaccine production if a mutated Hantavirus emerged. The scenario shows AI-biotech response times measured in days rather than years. No government delay would be tolerated.

Microsoft releases a completely free 15+ lesson course covering agentic RAG, multi-agent systems, tool use, memory, planning, and MCP/A2A protocols. The open-source curriculum includes code and videos for building real-world agents. It targets developers moving from zero to production.

A humanoid-robot-operated coffee shop in China runs with zero human staff using proprietary embodied AI. The model is expanding to 20+ locations across multiple cities with plans for 100+ nationwide. Coffee prices drop to ~$1.40 thanks to 24/7 operation and minimal labor costs.

Lun Wang leaves Google DeepMind and publishes a detailed critique arguing existing benchmarks cannot predict emergent capabilities or detect deceptive behavior in future models. He calls for self-evolving evaluation systems that co-evolve with models. The post highlights a critical blind spot in AI safety.

NVIDIA ships its first custom Vera CPUs to Anthropic, OpenAI, and SpaceX specifically optimized for agentic workloads. The move completes NVIDIA's push into the full AI compute stack beyond GPUs. Partners are already integrating the chips into production agent systems.

A single developer launches Halupedia, a Wikipedia-style site where every article is generated on the fly by AI and never existed before. The open-source 'vibeserver' repo powers a mirror universe of plausible but entirely fabricated knowledge. It has already gone viral for its creative absurdity.

Microsoft open-sources a 4B parameter model that converts any image into a textured 3D mesh in under 3 seconds using the new O-Voxel format. Outputs are ready for Blender, Unity, and Unreal with full PBR textures. The model runs efficiently on CUDA and is fully open source.

China's upcoming five-year plan includes 500,000 new AI positions plus massive vocational retraining and renewable-powered data centers. This stands in stark contrast to US approaches focused on private investment without coordinated workforce programs. The plan emphasizes carbon reduction targets.

A clear visual breakdown of agentic AI layers—Orchestration, Agents, Tools, Memory, Monitoring, Reliability, and Governance—helps developers design production systems. The post emphasizes that architecture, not just agents, determines real power. It has resonated widely with practitioners.

Google DeepMind teases breakthroughs and new tools ahead of tomorrow's Google I/O keynote. Leaks point to a massive new Gemini model and the 'Spark' personal agent. The event is expected to showcase significant multimodal and agentic advances.

Higgsfield's self-learning AI agent analyzes competitors 24/7, generates insights, and creates video ads with a single prompt. The cloud-native system uses 40+ tools and multi-layer memory accessible via browser or Telegram. Early users report replacing manual marketing workflows.

Nous Research's Hermes Agent brings reliable, self-improving agentic AI to NVIDIA RTX PCs and DGX Spark. The system runs locally with strong performance on consumer hardware. It targets developers building autonomous workflows on the desktop.

Government AI investments show US at $328B, China at $133B, and India at just $2B. Reports from Japan Times label India the 'biggest AI loser' as talent and capital flow elsewhere. The gap raises concerns about long-term competitiveness.

An IBM engineer breaks down the evolution from keyword search to agentic RAG in a concise video that has gone viral. The explanation covers intent understanding, external memory, and autonomous decision-making. It resonates across language communities.

Thousands of construction workers live in RVs parked for miles in Abilene, Texas, building massive AI data centers for OpenAI, Oracle, Microsoft, and others. Power demand will exceed one million homes and water use will reach millions of gallons daily. The boom is reshaping local economies.

New Gemini prompts turn the free model into a sophisticated stock analyst capable of replacing expensive Bloomberg terminals. The viral thread shares 10 ready-to-use prompts for deep financial analysis. Users are already applying them for real trading decisions.

Japan is described as missing the AI boat due to long-standing software challenges despite strong hardware materials supply. Analysts note the country risks alienating customers with poor AI features. The discussion ties into broader national competitiveness concerns.

Chatbase reaches $10M ARR and launches as a full harness for customer experience agents, providing context, tools, workflows, and guardrails. The platform positions itself as the 'Claude Code' equivalent for brand ambassadors. Next target is $100M ARR.

Pope Leo XIV is scheduled to meet Anthropic co-founder Christopher Olah next week to discuss artificial intelligence. The meeting highlights growing global interest in AI ethics and governance at the highest levels. It follows increasing Vatican engagement with tech leaders.

DA Davidson raises NVIDIA's price target to $300 citing continued AI compute leadership and ecosystem strength. The analyst notes demand is flowing back to NVIDIA platforms despite rising competition. The upgrade comes ahead of earnings and reflects sustained bullish sentiment.