Odaily Planet Daily News: In the early morning of May 15th, Beijing time, at the keynote speech of the Google I/O Developer Conference, Google CEO Sundar Pichai released dozens of Google and AI combined products, which can be called the "family bucket" level, and fought against OpenAI with all their strength, including Gemini 1.5 Pro and Gemini 1.5 Flash that support 2 million token long texts, Google version of Sora technology Veo, the strongest open source model Gemma 2, AI Overviews that support generative search, and the sixth generation TPU. But the most popular is the universal assistant project Project Astra, which is truly the way to AGI, announced by Google DeepMind CEO and Google AI head Demis Hassabis, and the voice NotebookLM, which directly targets GPT-4o. In the keynote speech of the conference, Google released a very amazing AI assistant demonstration, and this time it swore that it was not forged or tampered with in any way. Demis said that Project Astra is the prototype of the AI assistant that he has been looking forward to for decades, and it is the future of general AI. It is reported that Project Astra is a real-time, multimodal AI assistant that interacts with the world around it by receiving information, remembering what it has seen, processing that information, and understanding contextual details. Its voice interaction is more natural than the current form of Google Assistant, without delay or latency, and can answer questions or help you do almost anything. From the demonstration, Project Astra can process the information received very quickly. It can do this by continuously encoding video frames and combining video and voice into an event timeline, and then cache the information for recall. Google said that some of these features will be added to the Gemini application and other products later this year. In addition to Project Astra, Google has also launched a number of general AI Agent sub-series products based on Gemini, including audio NotebookLM, music Music AI Sandbox, video Veo, and image Imagen 3, which are directly benchmarked against GPT-4o, Dall-E, and Sora released by OpenAI. (Titanium Media)
