In 2016, developers at a Google subsidiary called DeepMind AI created the AlphaGo algorithm that made history by beating the champion in the Go board game. Now, Demis Hassabis, co-founder and CEO of DeepMind, announced the development of a neural network that will be more advanced than that behind Open AI’s popular ChatGPT AI bot.
The Gemini neural network is a large language model similar to the underlying ChatGPT-GPT-4 model. DeepMind developers intend to combine the capabilities of the neural network with the technologies used in AlphaGo to expand the algorithm’s capabilities. “At a high level, you can think of Gemini as combining some of the strengths of systems like AlphaGo with the amazing capabilities of large language models. We also have some innovations that will be very interesting.”– quotes the source of the word Demis Hassabis.
Recall that the development of the Gemini neural network first became known last month. As for AlphaGo, this algorithm was trained to solve complex problems in a way that required a specific action from the algorithm. In addition, an algorithm called tree search was used to examine and predict possible moves on the Go board. The Gemini algorithm is still in the development phase, which will take several months and could cost tens or even hundreds of millions of dollars to implement, Hassabis said. According to the available data, the development of GPT-4 cost more than $ 100 million.