In 2016, a robot called AlphaGo (Alpha Dog) instantly caused a sensation around the world by defeating Go world champion Li Shishi 4: 1 in a Go match.。Google's DeepMind artificial intelligence laboratory, which developed AlphaGo, has been controversial for a while.。Now, DeepMind says it will use AlphaGo to create a new artificial intelligence system.。

DeepMind co-founder and CEO Demis Hassabis said his engineers are using AlphaGo's technology to build an artificial intelligence system called "Gemini," which will be more powerful than the system behind OpenAI's ChatGPT.。

Gemini is still in development and will take months。According to Hassabis, Gemini is a large language model for processing text, essentially similar to GPT-4, which provides support for ChatGPT.。But DeepMind's team will combine the technology with that used in AlphaGo, aiming to give the system new capabilities, such as the ability to plan or solve problems.。

"At a high level, you can think of Gemini as combining some of the advantages of the AlphaGo system with the amazing language capabilities of the big model.。Hassabis said, "We also have some very interesting new innovations.。"

AlphaGo is based on a technique pioneered by DeepMind called "reinforcement learning," in which AlphaGo learns to deal with difficult problems, such as choosing what action to take in Go or video games, by repeatedly trying and receiving feedback on its performance.。In addition, it uses the technical method of Monte Carlo tree search to explore and remember possible actions on the chessboard。The next big leap for language models may involve them performing more tasks on the internet and computers。

Training large language models like OpenAI's GPT-4 requires a large selection of text from books, web pages, and other sources to be fed into machine learning software。It uses patterns in training data to skillfully predict which letters and words should appear after a piece of text.。It turns out that such a simple mechanism is very powerful in answering questions and generating text or code。

An important additional step in making ChatGPT and similar functional language models is to use reinforcement learning based on human feedback on AI model answers to improve their performance。DeepMind's extensive experience in reinforcement learning could allow its researchers to give Gemini novel abilities。

Hassabis and his team may also try to use ideas from other areas of artificial intelligence to enhance language big model technology.。The DeepMind researchers' research spans fields ranging from robotics to neuroscience, and earlier this week the company demonstrated an algorithm capable of learning to perform operational tasks using a variety of different robotic arms.。

Developing Gemini could cost tens of millions or hundreds of millions of dollars.。By comparison, OpenAI CEO Sam Altman revealed in April that it cost more than $100 million to create GPT-4.。

Previously, after the stunning appearance of ChatGPT, Google, although also launched its own chat robot "Apprentice Bud" (Bard), and the application of generative artificial intelligence to its search engine and many other products, but inevitably still feel a step too late。And once Gemini is completed and launched, it will become an important driver of Google's "charge" in the face of ChatGPT and other AI technology competitors.。

In an effort to bolster AI research, Google in April merged its main AI lab, Google Brain, with Hassabis' unit DeepMind, eventually merging into a new unit called Google DeepMind.。Hassabis said the new team brings together two powerful forces that have laid the foundation for recent advances in artificial intelligence.。"If you look at the progress we've made in AI, I think 80 or 90 percent of the innovation has come from one of them," Hassabis said.。"

In 2014, DeepMind was acquired by Google after demonstrating the amazing results achieved by software that uses reinforcement learning to master simple video games.。Over the next few years, DeepMind showed how the technology could accomplish what once seemed uniquely human.。In particular, when AlphaGo defeated Go champion Li Shishi in 2016, many AI experts were shocked because they thought it would take decades for robots to master such a complex game.。

When ChatGPT came out, many experts expressed concern about the development of artificial intelligence。Rapid advances in language models have many AI experts - including some who build algorithms - worried about whether the technology could be used maliciously or become unmanageable.。Some tech insiders have even called for a moratorium on developing more powerful algorithms to avoid producing something dangerous.。

Earlier, thousands of tech people jointly issued an open letter calling for the suspension of training for AI systems more powerful than GPT-4.。In an open letter, the tech leaders called on all AI labs to suspend training on AI systems more powerful than GPT-4 for at least the next six months.。Otherwise, the government should step in and take coercive measures.。

Business people such as Tesla's Musk and Apple co-founder Steve Wozniak are also in the open letter.。

Subsequently, Bill Gates came out to make a public statement, saying that suspending AI development would not really solve the problem.。Bill Gates said that asking a specific group to suspend the development of artificial intelligence technology will not solve the related challenges.。Obviously, these things are practical。What we need to do is identify the tough issues and fix them。

Hassabis and Bill Gates have similar views。He said artificial intelligence has huge potential benefits, such as scientific discoveries in areas such as health or climate, so humans must continue to develop this technology.。He also argued that a mandatory moratorium would be impractical as it would be almost impossible to enforce。"If handled properly, this will be the most beneficial technology ever to humanity," he said of artificial intelligence.。"

Hassabis is currently tasked with managing unknown and potentially serious risks in addition to accelerating Google's AI development.。DeepMind had been exploring the potential risks of AI long before the advent of ChatGPT, and one of the company's co-founders, Shane Legg, has been leading the company's internal "AI security" group for years.。Hassabis signed a statement last month along with other high-profile AI experts warning that AI could one day pose risks comparable to nuclear war or epidemics.。