Googles AI Boss Reveals What AI In 2026 Looks Like

Future of AI: Insights from Demis' Interview

Overview of AI Developments by 2026

• Demis discusses his vision for the future of AI, particularly focusing on the emergence of full omnimodels by 2026.

• He emphasizes the convergence of modalities, highlighting Google's Gemini as a foundational multimodal model capable of processing images, video, text, and audio.

• The advancements in multimodality are exemplified by Google’s latest image model, Nano Banana Pro, which demonstrates impressive visual understanding and infographic creation capabilities.

Key Components of Full Omnimodel Stack

• The full omnimodel stack consists of six components: robotics, images, video, audio, 3D, and text.

• Google is rapidly advancing in these areas; however, it is noted that they are slightly behind in robotics compared to other frontier AI companies.

Innovations in Robotics with Gemini Robotics 1.5

• Introduction of Gemini Robotics 1.5 aims to enhance physical agents' capabilities to solve complex multi-step challenges effectively.

• Demonstrations show robots can perform tasks like sorting fruits or laundry through step-by-step reasoning and environmental perception.

• The new model allows for uniformity across different robot types without needing fine-tuning for various form factors.

Advanced Capabilities and Applications

• Gemini Robotics 1.5 can utilize internet resources to answer questions and complete tasks based on specific guidelines (e.g., waste sorting).

• This advancement signifies a move towards creating genuinely useful AI agents that can assist in everyday physical tasks.

Multimodal Integration: Images and Videos

• The integration between image generation (Nano Banana Pro) and potential applications in video showcases how models reason like agents during content creation.

• Future developments may extend this reasoning capability into video production as well.

V3 Video Technology

• V3 is highlighted as a leading technology for image-to-video conversion with expectations for significant improvements by 2026.

Underappreciated Features: Gemini Live

• Discussion includes Gemini Live as an underrated feature combining live speech recognition with real-time reasoning capabilities to assist users effectively.

AI-Assisted Oil Change Demonstration

Overview of AI in Practical Applications

• The demonstration showcases an individual using Gemini Live to perform an oil change on a 2009 BMW 335i, highlighting the practical application of AI in everyday tasks.

• The speaker speculates about advancements by 2026, predicting improvements in latency and reasoning capabilities of AI systems, enhancing their utility for complex tasks.

Step-by-Step Oil Change Process with AI Guidance

• Gemini assists the user by asking about the type of oil and tools available, demonstrating interactive communication between AI and user.

• The AI provides specific instructions on locating the oil filter and drain plug, emphasizing its role as a guide through technical processes.

• Instructions include removing a plastic panel to access the drain plug, showcasing how AI can simplify mechanical tasks for users.

Technical Specifications and User Interaction

• The conversation includes torque specifications for reassembling parts (18 ft-lb), illustrating how detailed guidance is provided by the AI during each step.

• Users are prompted to replace O-rings on the oil filter cap before installation, indicating thoroughness in maintenance procedures facilitated by AI.

Completion and Future Implications

• After completing the oil change process, Gemini confirms that everything is set, reflecting successful interaction between human and machine.

• The speaker expresses optimism about future developments in AI technology by 2026, suggesting significant enhancements in user experience and task execution.

Exploring World Models: A New Frontier

Introduction to World Models Concept

• Discussion shifts towards "world models," which are anticipated to be a major theme in technological advancements by 2026.

Genie3: Interactive Video Model

• Introduction of Genie3 as an innovative system allowing users to generate interactive video environments that respond dynamically to user actions.

Real-Time Interactivity Features

• Genie3 enables real-time interactivity where environments react live based on user movements rather than relying on pre-built simulations.

• The concept of "world memory" is introduced; actions taken within these environments persist over time, enhancing immersion.

Dynamic Content Creation

• Users can create new events within their generated worlds spontaneously (promptable events), showcasing flexibility and creativity afforded by advanced world models.

Genie 3: Revolutionizing World Simulation

Exploring Real-World Physics and Unique Environments

• Genie allows users to explore real-world physics, movement, and various unique environments, including distinct geographies and fictional settings.

• The application of Genie 3 extends to next-generation gaming and entertainment, as well as research in embodied training for robotic agents.

Advancements in World Models

• The transition from Genie 2 to Genie 3 showcases significant advancements in world models that could lead to enhanced simulations powered by cross-modality.

• These world models are characterized by their ability to simulate virtual worlds with memory and reasoning capabilities, which were previously unimaginable.

Agent-Based Systems and Their Potential

• Google is leading the development of agent-based systems that aim to perform full tasks reliably; however, they still face challenges in reliability.

• Gemini 2.0 O powers a multi-agent system designed as a virtual scientific collaborator that assists researchers in generating novel testable hypotheses.

Future of AI Agents in Scientific Research

• The potential for AI agents to propose new ideas and conduct experiments marks a transformative shift in scientific research methodologies.

• Google's code men agent focuses on detecting and fixing security vulnerabilities within codebases, showcasing the versatility of AI applications across different fields.

Comprehensive Data Science Automation

• Google's data science agent automates end-to-end data science workflows within Google Collab, highlighting the extensive range of tools being developed under their roadmap.

• Alpha Evolve represents another innovative step forward as an AI scientist focused on algorithmic discovery, indicating the future trajectory of AI's role in scientific exploration.