DeepMind next-generation model will combine AlphaGo to surpass ChatGPT

DeepMind next-generation model will combine AlphaGo to surpass ChatGPT

DeepMind next-generation model will combine AlphaGo to surpass ChatGPT

Demis Hassabis, CEO of Google’s research lab DeepMind, told Wired that its engineers are using AlphaGo technology to develop an artificial intelligence system called Gemini that is more powerful than the system behind OpenAI ChatGPT.

Gemini is essentially similar to GPT-4, a large language model that can process text; it is still under development, it will probably take a few months, and it is expected to cost tens or hundreds of millions of dollars. OpenAI CEO Sam Altman revealed in April that the cost of creating GPT-4 exceeded $10 billion.

Hassabis said his team plans to combine the technique with that used in AlphaGo, with the aim of giving the model new capabilities, such as planning or problem solving. “At a high level, you can think of Gemini as combining some of the advantages of an AlphaGo-like system with the amazing language capabilities of large models. We also have some really interesting innovations.”

Based on reinforcement learning techniques pioneered by DeepMind, AlphaGo learns to deal with thorny problems that require choosing which action to take, such as the game of Go or video games, by having it try repeatedly and receive feedback on its performance.

It also uses a method called Monte Carlo tree search to explore and memorize possible moves on the board.

In 2016, AlphaGo became famous by defeating the world Go champion Lee Sedol 4-1, becoming the first artificial intelligence robot in history to defeat a human professional Go player and the Go world champion.

“Wired” magazine believes that the next big leap in language models may involve them performing more tasks on the Internet and computers. When Gemini is complete, it could play an important role for Google in addressing the competitive threat posed by ChatGPT and other generative AI technologies.

In fact, Gemini isn’t DeepMind’s first foray into language models; the company launched a chatbot called Sparrow last year.

Hassabis revealed in January that they considered launching a private beta of Sparrow sometime this year; it’s unclear if those plans are still on track.

In addition to driving Google’s AI efforts forward, Hassabis also has the responsibility of managing unknown and potentially serious risks.

In response to some recent concerns about the development of AI, Hassabis believes that AI has extraordinary potential benefits, such as scientific discoveries in fields such as health or climate, so that humans must continue to develop this technology.

Mandatory suspension is impractical and nearly impossible to enforce. If done right, AI will be the technology that will benefit humanity most, and “we have to be bold and courageous in our pursuit of these.”

But that doesn’t mean Hassabis is advocating the development of AI to be herded. He joined other prominent AI industry experts in signing a statement last month warning that AI could one day pose risks comparable to nuclear war or a pandemic.

One of the biggest challenges right now, Hassabis noted, is determining what the risks of more capable AI might be. He believes that more research, evaluation tests, etc. should be done in the field to determine the capabilities and controllability of new artificial intelligence models.

To that end, DeepMind may make its systems more accessible to outside scientists ; “I’d like to see the academic community get early access to these cutting-edge models.”

DeepMind next-generation model will combine AlphaGo to surpass ChatGPT