Shulou(Shulou.com)11/24 Report--

Google DeepMind CEO Hassabis new revelation: the new Gemini model will be combined with AlphaGo and big language model, the cost is estimated to be tens of millions of dollars, even hundreds of millions of dollars.

Google is really burning its bridges.

Is Gemini, the legendary merger of AlphaGo and GPT-4-like models, finally coming?

One is the AI system that defeats the human go champion and makes history with reinforcement learning, and the other is the list of almost all the big models, the strongest multimodal large model, and the combination of two AI, which is almost invincible!

Google DeepMind CEO Hassabis recently told foreign media Wired that Gemini is still under development and will take several months, while Google DeepMind is ready to spend tens of millions of dollars, even hundreds of millions of dollars.

Sam Altman has previously revealed that the cost of creating GPT-4 is more than $100 million. Google DeepMind, of course, cannot lose.

If the Gemini is too long, it will merge the language functions of large models such as AlphaGo and GPT-4, and the system's ability to solve problems and plan will be greatly enhanced.

Gemini is a large language model, similar to GPT-4

It is expected to cost tens to hundreds of millions of dollars, which is comparable to the cost of developing GPT-4.

Besides AlphaGo, there will be other innovations.

Gemini will integrate AlphaGO with reinforcement learning and tree search.

Reinforcement learning allows AI to solve challenging problems by learning from repeated attempts and feedback

The tree search method helps to explore and remember possible movements in a scene, such as in a game scene

DeepMind's rich experience in reinforcement learning will bring new features to Gemini.

Other areas of technology (such as robotics and neuroscience) will also be integrated into Gemini

The next algorithm is to surpass ChatGPT. According to OpenAI CEO Sam Altman, GPT-5 is still some time away from release and will not start training for at least six months. The release date of the Gemini has not yet been set, possibly within a few months.

Gemini, which is still under development, is also a large language model for dealing with text, similar in nature to GPT-4.

But Google DeepMind CEO Demis Hassabis says Gemini will incorporate the technologies used in AlphaGo, which will give the system a whole new ability to plan and solve problems.

The scene of AlphaGo beating go world champion Lee se-dol in 2016 is still fresh in my mind.

Hassabis said, "it's fair to say that Gemini combines some of the advantages of the AlphaGo system with the amazing language capabilities of the large language model. And we have some other interesting innovations. "

It is said that Gemini has multimodal capabilities that were not found in previous models and is very efficient in integrating tools and API. Moreover, Gemini will be available in a variety of sizes designed to support future memory and planning innovations.

In March, there was a saying that Gemini, like GPT-4, would have a trillion parameters. Moreover, it is said that Gemini will use tens of thousands of Google TPU AI chips for training.

At last month's Google developer I / O conference, Google mentioned that from the beginning, the goal of Gemini is multimodal, efficient integration tools, API.

At the time, Google's prediction was: "although it is still early, but we have seen in Gemini, never seen in the previous model of multimodal capabilities, which is very impressive." "

The technology behind AlphaGo is reinforcement learning, which is pioneered by DeepMind.

RL agents interact with the environment over time and learn strategies through trial and error to maximize long-term cumulative rewards. Through reinforcement learning, AI can adjust its performance through trial and error and receive feedback, thus learning to deal with thorny problems, such as choosing how to take the next step in go or video games.

In addition, AlphaGo uses the Monte Carlo Tree search (MCTS) method to explore and remember all possible actions on the chessboard.

This is not the first time Hassabis has stirred the tech giant's massive artificial intelligence gold rush.

In 2014, DeepMind used reinforcement learning to teach AI to play simple video games, an amazing achievement that led to the acquisition of DeepMind by Google.

Google made the right bet.

Over the next few years, DeepMind produced an amazing result from time to time.

Deep learning and reinforcement learning are solving many classic artificial intelligence problems, such as logic, reasoning and knowledge representation. In 2016, the earth-shattering AlphaGo directly ignited the boom of deep learning and the first round of the AI industry.

In 2017, AlphaGo Zero quickly overtook AlphaGo without using human data.

In the year of AlphaGo Zero2020, AlphaFold's prediction of protein structure was similar to that of laboratory technology, which basically solved the problem of protein folding.

In June this year, AlphaDev created a whole new sorting algorithm that could revolutionize the efficiency and results of computer science.

Compared to OpenAI's more general route, DeepMind has ploughed deep into the vertical field for many years.

Where is the next big leap forward in the language model? Gemini may point the way for the next generation of language models.

It is clear that Gemini is a back-to-back battle for Google.

Many of the technologies pioneered by Google, such as the Transformer architecture, make the recent flood of AI possible.

Because it is too cautious in the development and deployment of technology, it lags behind temporarily in the face of competition from ChatGPT and other generative AI.

In order to fight ChatGPT, Google has thrown out a number of actions in succession, such as launching Bard and integrating generative AI into search engines and other products.

In order to focus on big things, Google simply merged Hassabis's DeepMind with Google brain, Google's main artificial intelligence laboratory, into Google DeepMind in April.

For the new team after integration, Haasabis is obviously very confident. He says the new team brings together two forces that are crucial to recent advances in artificial intelligence.

"if you look at our position in the labor field, you will believe that 80% or 90% of the innovation in the future will come from one of the teams. Over the past decade, both teams have achieved excellent results. "

New ideas train large language models like OpenAI's GPT-4, which requires the input of a large number of selected data sets from books, web pages, and other sources into "Transformer."

Transformer uses patterns in the training data to skillfully predict every letter and word that should appear in subsequent text.

This seemingly simple mechanism is very powerful in answering questions and generating text or code.

But this seemingly simple technical principle has also been criticized by many industry leaders or artificial intelligence experts.

Musk: the essence of current AI technology is statistics.

LeCun: the current level of intelligence of AI is not as good as the breakthrough made by OpenAI in the series of GPT models, which is based on the core technology of Transformer and radically uses RLHF to strengthen the ability of the model.

DeepMind also has a wealth of experience in reinforcement learning.

This gives people every reason to expect the innovative capabilities that Gemini may demonstrate in the future.

More crucially, Hassabis and his team will also try to enhance the capabilities of large language models with core technologies in other areas of artificial intelligence.

The technology accumulation of DeepMind is very extensive.

From robotics to neuroscience, they have a variety of equipment to choose from in their arsenal.

AI bigwigs like LeCun, for example, say that Transformer overlimits the power of language models to the scope of text.

Like humans and animals, learning from the world's physical experience may be the best solution for the development of artificial intelligence.

Perhaps in Gemini, artificial intelligence will show potential in other directions.

The task of Hassabis in the uncertain future is to accelerate the development of Google's artificial intelligence technology while managing unknown and potentially serious risks.

The rapid progress of the large language model has made many artificial intelligence experts worry about whether this technology will open Pandora's magic box and make human society pay an unacceptable price.

Hassabis said that the benefits that artificial intelligence may bring to human society are incalculable.

Mankind must continue to develop this technology.

Forcing a moratorium on the development of AI technology is not operable at all.

But that doesn't mean that Hassabis and his DeepMind will push technology forward regardless of the consequences.

After all, Google and DeepMind ceded the leadership of AI technology to OpenAI.

A large part of the reason is the "overly responsible" attitude towards the development of AI.

Netizens: not optimistic, but for the future release of Gemini, because considering Google's conservative attitude before, most netizens do not seem to be optimistic.

When do you think this AGI-like model will be released?

I bet $10 Google will never release this thing.

If anyone has followed Google's project, they will find that they usually brag for a while, then release nothing, and then cut the project a year later.

However, netizens still agree with Google's contribution to the current big language model.

The big language modeling technology used by netizen A:OpenAI is basically invented by Google.

Netizen B: yes, but Tesla can't get rich, but Edison can.

This netizen is optimistic that DeepMind will use his experience in reinforcement learning to make a breakthrough in the big language model.

But he still thinks it's possible that Google will only promote the technology by improving its existing products, rather than launching entirely new ones.

Reference:

Https://the-decoder.com/deepmind-founder-shares-details-on-gemini-googles-next-gen-response-to-gpt-4/

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.