๐ ThursdAI - LIVE from AI Engineer Worlds Fair - OpenAI, DeepMind, EXO, Sakana & more friends
Summary
The AI Engineer World's Fair 2026 in San Francisco, attended by 7,200 people, showcased rapid advancements and evolving challenges in AI. Key discussions included the return of OpenAI's Fable model after a US government ban, alongside its new Soul, Terra, and Luna models, and Codex's internal adoption with features like token banking and sub-agents. Google DeepMind unveiled Gemini Omni Flash for video generation/editing and Nano Banana 2 Light for fast image generation. Exolabs introduced local.ai and EXO CLI to enable local AI on consumer devices, emphasizing "freedom of intelligence." Sakana AI presented Fugu, a router model for optimal task-specific model selection, based on evolutionary and reinforcement learning. Weights & Biases launched Arya, an agent for automated research. The conference also highlighted the "token billionaire" concept and plans for global expansion to Tokyo and Tel Aviv.
Key takeaway
For AI Engineers navigating the rapidly evolving landscape, prioritize exploring local AI solutions like Exolabs' EXO CLI for sovereignty and cost efficiency, especially given recent model access restrictions. Evaluate specialized models such as Sakana AI's Fugu for task-specific orchestration and Google DeepMind's Omni Flash for multimodal applications. Embrace agentic workflows with tools like OpenAI's Codex and W&B's Arya, but maintain human oversight for critical code and decision-making.
Key insights
AI's rapid evolution necessitates local, agentic, and specialized models for sovereignty and optimal performance.
Principles
- Local AI ensures freedom of intelligence against external restrictions.
- Model routing optimizes performance by selecting task-specific models.
- Human responsibility remains crucial for AI-generated code and decisions.
Method
Exolabs' EXO CLI simplifies local VLM deployment on consumer devices with optimal configurations. Sakana AI's Fugu uses evolutionary and reinforcement learning for recursive prompt rewriting and output verification.
In practice
- Use EXO CLI for local VLM deployment on MacBooks or DGX Spark.
- Leverage Fugu for robot control, typhoon prediction, or GPU optimization.
- Employ W&B Arya for automated loss curve analysis or gradient issue detection.
Topics
- AI Engineer World's Fair
- Local AI
- Large Language Models
- AI Agents
- Multimodal AI
- Model Routers
- AI Hardware
Best for: AI Architect, NLP Engineer, Computer Vision Engineer, AI Engineer, Machine Learning Engineer, AI Scientist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Weights & Biases.