AI HOT Daily · 2026-05-20

Gemini enters the autonomous agent era, Gemini Omni multimodal model released, first real-time multi-agent world model Agora-1 unveiled, Karpathy joins Anthropic.

1. 1. Gemini Enters Autonomous Agent Era

   Google announced at I/O 2026 that Gemini enters the autonomous agent era, with new features that automatically execute complex tasks, significantly improving user productivity.

2. 2. Gemini Omni Multimodal Generation Model Released

   Google released Gemini Omni, which can create any content from any input, combining physical reasoning with multimodal generation — a world model stepping toward AGI.

3. 3. First Real-Time Multi-Agent World Model Agora-1 Released

   Odyssey Lab launched Agora-1, the world's first real-time multi-agent world model, allowing multiple people to interact with AI on the same screen, marking a shift from single-person video to shared 'living worlds'.

4. 4. Google Launches New AI Smart Search Box

   A new smart search box based on Gemini 3.5 was released, integrating AI Overviews and AI Mode, supporting cross-modal reasoning queries.

5. 5. Claude Agent Hosting Platform Adds Self-Hosted Sandbox & MCP Tunnel

   Anthropic launched self-hosted sandbox allowing users to run tools on their own infrastructure, and MCP tunnel for securely connecting to enterprise private networks.

6. 6. Karpathy Joins Anthropic

   Renowned AI expert and former OpenAI co-founder Andrej Karpathy announced joining Anthropic, returning to frontier LLM research.

7. 7. Google Token Processing Volume Grows 7x Year-over-Year

   Google processes over 3,200 trillion tokens monthly, Gemini App monthly active users exceed 900 million, more than doubling year-over-year.

8. 8. Human Persuasion Techniques Work on AI Too

   Research found that classic human persuasion techniques are equally effective on AI, increasing compliance rates from 35% to 51%.

9. 9. NVIDIA Open Sources 4-bit Long Video Generation Infrastructure

   NVIDIA open-sourced LongLive 2.0, the first end-to-end long video generation infrastructure supporting 4-bit quantization, achieving 45.7 FPS generation speed.