The Google-owned DeepMind laboratory is developing a new AI model, Gemini, expected to surpass OpenAI’s ChatGPT. According to Wired заявил Demis Hassabis, CEO of the company.
According to Hassabis, the technology uses the AlphaGo algorithm, based on reinforcement learning. The neural network is similar in function to its competitor, but will also be able to plan for the future and make decisions.
“At a high level, you can think of Gemini as a fusion of some strengths of systems like AlphaGo with the impressive language capabilities of large models. We also have a few new innovations, quite interesting,” added Hassabis.
He explained that Gemini will be in development for several more months. The project is valued at ‘tens or hundreds of millions of dollars’.
The head of DeepMind noted the general benefits of AI for research in health and climate. However, Hassabis emphasised that the main challenge at present is assessing the potential risks of more powerful artificial intelligence.
“If everything is done right, this will be the most useful technology for humanity. We must boldly pursue our goal,” the developer concluded.
Earlier, in January 2022, cofounder Mustafa Suleyman ушел из Google.
In February 2023, the company представила Bard chatbot. A month and a half later, the firm открыла доступ to a limited number of users to the tool.
In April, Google’s leadership объявило the merger of its AI divisions Brain and DeepMind.