Google's DeepMind is developing an alternative to ChatGPT in its labs. Dubbed the world's most sophisticated AI model, it will leverage strengths from AlphaGo. It will play a key role in the counterattack being mounted by Google against OpenAI and Microsoft.

DeepMind, Google's wholly owned artificial intelligence subsidiary, is currently working on an alternative to GPT-4, the latest iteration of the linguistic model underlying ChatGPT. Named Gemini, this new AI uniquely combines the work of a linguistic model with past technologies developed by the seasoned AI company. Like GPT-4, Gemini is multimodal. This means the AI can process multiple data sources like text, audio or images. In a Wired interview, Demis Hassabis, co-founder and CEO of DeepMind, revealed new details about the project still in development. Per Hassabis, DeepMind's AI will become the most advanced language model in the world, surpassing OpenAI's ChatGPT. For reference, Google mentioned Gemini last month at its Google I/O conference dedicated to Bard and PaLM 2.

To develop its linguistic model, Deepmind, acquired by Google in 2014, leveraged achievements from AlphaGo, the artificial intelligence it developed itself. Unveiled in 2016, this AI caused great excitement by defeating several champions at Go - a strategic game popular in Asia. But after several victories, the AI lost to a Korean player. "Gemini combines some of the strengths of AlphaGo-like systems with the amazing linguistic capabilities of large models," explains Demis Hassabis, promising "new innovations that will be quite interesting." Thanks to AlphaGo, Gemini will be able to plan tasks and solve problems efficiently.

To complete work on Gemini, Google decided to combine Deepmind with teams from Brain, a deep learning research project launched in 2011. With this reorganization, the search giant is evidently trying to catch up to its competitors, OpenAI and Microsoft. According to BR24, a German radio station, some engineers from the startup recently left for Google. Per Demis Hassabis, Gemini still has several months of development before being fully operational. Once refined, the AI is meant to power most of Google's AI-related services, including the Bard chatbot and the company's future search system. Therefore, Gemini is poised to succeed PaLM 2.