At this year's Google I/O Developer Conference, Google announced a series of new advancements in AI, including Project Astra - an ambitious plan to build a future general AI agent. During the conference, Google showcased an early version of Project Astra, which aims to build a multimodal AI assistant that can observe and understand the dynamic world and provide real-time responses to help users with daily tasks or answer questions. This idea is similar to OpenAI's recently demonstrated ChatGPT powered by GPT-4o. However, as GPT-4o is set to be launched for ChatGPT Plus subscribers in the coming weeks, Google's actions in this regard seem slightly lagging. Currently, Google is still conducting in-depth research on the Astra project and has not revealed when the mature AI agent will be officially released. However, the company stated that some features of Project Astra will first appear in the Gemini assistant later this year. What new features will Project Astra bring? Based on the advancements in Gemini Pro 1.5 and other specific task models, Project Astra, the advanced responsive agent for seeing and speaking, will allow users to interact with an AI assistant while sharing complex dynamic surroundings. This assistant can understand what it sees and hears and provide accurate responses in real-time. Demis Hassabis, CEO of Google DeepMind, wrote in a blog post, "To be truly useful to users, AI agents need to understand and respond to the complex and ever-changing world like humans - they must be able to absorb and remember everything they see and hear to understand context and take action. Additionally, they need to be proactive, trainable, and personalized, so users can have natural conversations without worrying about delays or procrastination." In a demo video released by Google, a prototype Project Astra agent running on a Pixel smartphone can recognize objects, describe their specific components, and understand code written on a whiteboard. Even more astonishingly, it can identify neighbors through the camera viewfinder and demonstrate its memory capabilities by telling users where they put their glasses. Another demo video shows a scenario where the agent suggests improvements to the system architecture while overlaying real-time images of a pair of glasses in the user's field of view. Hassabis pointed out that although Google has made significant progress in reasoning across multimodal inputs, reducing the agent's response time to a level close to human conversation is still a challenge. To address this issue, Google's agent processes information by continuously encoding video frames, merging video and voice inputs into event timelines, and caching this information for efficient recall. He added, "By leveraging our leading speech models, we have also enhanced the agents' vocal performances, giving them a wider range of intonations. These agents can better understand the context they are in and respond to conversations quickly." It is worth noting that OpenAI did not use multiple models in GPT-4o but trained the model end-to-end, covering text, vision, and audio, enabling it to handle all inputs and outputs with an average response time of 320 milliseconds. Google has not yet disclosed the specific response time of Astra, but as the project progresses, any potential latency is expected to be reduced. It is currently unclear whether the Project Astra agent possesses the same level of emotional understanding as GPT-4o. When will it be available? Although Google has not announced the specific release date for Project Astra, the company stated that some features of the project will first appear in the Gemini assistant later this year. Gemini is an intelligent assistant application launched by Google, which already has rich functionality and a good user reputation. With the addition of Project Astra, Gemini will become even more powerful, providing users with more comprehensive and intelligent services. During the conference, Google also revealed its plans to update Gemini Live. This feature will allow users to have two-way conversations with chatbots and discuss their surroundings by opening the camera. This feature will enable users to interact with AI assistants more naturally, further enhancing the user experience. Industry experts generally believe that Google's Project Astra will be a significant breakthrough in the field of AI. It will not only drive the development and application of AI technology but also bring more convenience and efficiency to users' lives. With the continuous progress and improvement of the project, we have reason to believe that future AI agents will become an indispensable part of our lives. The convening of the Google I/O conference marks the continuous investment and innovation of technology giants in the field of AI. Through the launch of projects like Project Astra, Google will continue to lead the development direction of AI technology and bring more convenience and well-being to human society.