Google I/O: LLM capabilities power agentic AI search | Computer ...
Google has made significant advancements in artificial intelligence (AI) language models to create what it refers to as "world models", aiming to enhance their utility and universality. The company unveiled the Gemini 2.5 large language model (LLM), along with new APIs, programming tools, and agentic AI features integrated into Google's search engine at its annual developer event, Google I/O.
Gemini and Other AI Models
Gemini serves as Google's primary AI engine, complemented by other models such as Gemma 3n, a smaller language model designed for mobile devices.
AI Transformation Vision
Demis Hassabis, the CEO of Google Deepmind, expressed the company's vision to evolve the Gemini app into a universal AI assistant capable of executing daily tasks, managing administrative duties, and offering innovative recommendations to enhance productivity and enrich lives.
Development of New AI Capabilities
Google has been exploring new AI capabilities, building on the groundwork laid by Project Astra, a research prototype focused on video comprehension, screen sharing, and memory functions. These capabilities have been integrated into Gemini Live for broader accessibility.
Advancements in AI Models
Google has been striving to transform its main AI model, Gemini, into a world model. With Gemini 2.5 Pro, the model has gained the ability to formulate plans and simulate novel experiences by comprehending various aspects of the world.
Enhanced AI Features
One of the features powered by Gemini 2.5 is AI Mode, now available on Google's North American search platform, offering deeper search capabilities compared to the existing AI Overview feature. Project Mariner, an agentic AI component embedded in AI Mode, aims to expedite online tasks for users, such as finding affordable tickets across multiple sites efficiently.
Support for Developers
Google has integrated Gemini 2.5 Pro into the native code editor of Google AI Studio to facilitate faster prototyping for software developers. Additionally, the beta version of Jules, an asynchronous code agent, has been introduced to collaborate directly with developers' GitHub repositories.
Conclusion
As Google continues to enhance its AI capabilities and explore new horizons in AI development, the integration of LLM capabilities and agentic AI features into Google's services marks a significant step towards a more efficient and intuitive user experience.
