Google’s CEO of DeepMind has announced the development of their next algorithm, Gemini, which aims to surpass the capabilities of the chatbot GPT. Leveraging techniques used in their previous AI, AlphaGo, Gemini combines language capabilities with large models like GPT-4 to enhance problem-solving and planning abilities. This advanced AI model is currently under development and is expected to cost tens to hundreds of millions of dollars, comparable to the cost of developing GPT-4. DeepMind also plans to implement new innovations in Gemini, including reinforcement learning and tree search methods. These techniques, previously used in AlphaGo, involve learning from repeated attempts and feedback, allowing the software to tackle challenging problems. Gemini’s success could significantly contribute to Google’s competitive position in the field of generative AI technology. Google has already integrated generative AI into its search engine and other products, and Gemini is part of their response to competitive threats posed by chatGPT and other generative AI technologies. With DeepMind’s extensive experience in reinforcement learning, Gemini is expected to have novel capabilities. The training process for Gemini involves predicting sequences of letters and words, and DeepMind is also exploring the integration of ideas from other areas of AI, such as Robotics and Neuroscience. Gemini is poised to be a breakthrough in the field of AI, and its development reflects Google’s commitment to pushing the boundaries of technology. Source: Wired
