Deep Mind wants to use AlphaGo technology to create a system that outperforms ChatGPT

Deep Mind wants to use AlphaGo technology to create a system that outperforms ChatGPT

28.6.2023

Demis Hassabis, co-founder and CEO of DeepMind, has seen opportunities in AlphaGo to evolve into a system with capabilities surpassing those of competitor ChatGPD.

“Gemini,” as the project is called, is a large language model that works with text and is similar to GPT-4. His team will combine this technology with the techniques used in AlphaGo, so that, for example, new skills such as planning or problem solving would be possible.

“At a high level,” says Hassabis, “you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the larger models.” He also didn’t forget to point out that they “also have some new innovations that will be very interesting”. What he means emerges from the following background

AlphaGo was based on a technique developed by DeepMind called “Reinforcement Learning” in which software learns to tackle difficult problems. As in Go or video games, in which a choice of actions to be taken is necessary, making repeated attempts and receiving feedback on their performance. A method called tree searching was also used to explore and remember possible moves on the board. The next big leap for language models could be that they perform more tasks on the Internet and computers.

The development of Gemini will take a few more months and cost hundreds of millions of dollars. Sam Altman, CEO of Open AI, recently mentioned the dimensions involved. The research work for GPT-4 alone would have cost over 100 million US dollars.

Gemini, of course, has one goal: to overtake Open AI/Microsoft’s competitor ChatGPT, as well as other generative AI technologies. Alphabet thought that this goal could be achieved with the chatbot Bard. But Google had to say goodbye to that after a short comparison. To fuel AI research on a massive scale, in April the company merged Hassabi’s DeepMind unit with Google’s primary AI lab, Brain, to create Google DeepMind.

According to Hassabis, this brought together two powerhouses that have been fundamental to recent AI advances. “If you look at where we are in AI, I would say that 80 or 90 percent of innovation comes from one or the other,” says Hassabis. “Both organizations have done great things over the past decade.

This could now increase, as Hassabis and his team could try to improve the technology of large language models with ideas from other areas of AI. DeepMind researchers work in fields ranging from robotics to neuroscience, and recently presented an algorithm that can learn to perform manipulation tasks with a variety of different robotic arms.

With the battle for supremacy in the generative AI space attracting more investors than usual, Grand View Research set out to quantify the prospects for the market. Result: By 2030, the market for generative AI — including text-analyzing AI like Gemini — could reach over $109 billion by 2030, a 35.6% increase from 2030.

Hits: 4