The founder of Google officially returns to join the LLM battle! Instructing Gemini R&D, the melee with OpenAI and Meta is on the verge of breaking out

Article source: Xinzhiyuan

Today, Silicon Valley once again staged a shocking drama. Sergey Brin returned to Google’s trenches, not only personally guiding Gemini’s research and development, but also directing the company to dig people out. How will the founders and veterans stir up the situation of the AI war?

The co-founder of Google, who has been resigned for four years, is finally back!

According to the Wall Street Journal, Sergey Brin has returned to work helping Google’s AI researchers build the powerful Gemini system.

Co-founder, returning to the big model battlefield

In recent months, billionaire Sergey Brin has been visiting Google’s Mountain View, Calif., offices three to four days a week to work with researchers pushing Google’s next big AI system.

In 2019, after Sergey Brin resigned from Google’s parent company Alphabet, he almost completely delegated power and rarely interfered in the company’s business.

But since the end of last year, he began to participate in Google’s conferences on AI many times, and the frequency has become significantly higher.

According to reports, regarding Google’s long-awaited AI model Gemini, he is working closely with researchers to study issues such as the “loss curve” of AI programs.

People familiar with the matter said that Sergey Brin even intervened in personnel issues, caring about the talent war and urging Google to hire the hottest researchers.

After the developer conference, the stock price has risen all the way

Although Google has done many pioneering AI technologies, it is slower to transform into new products than its competitors.

The competition in AI research is heating up, and Meta’s recent launch of Llama 2 is even more impressive.

Google plans to build a general-purpose AI program Gemini, which is said to be able to fight GPT-4.

Demis Hassabis, the Google executive in charge of the Gemini project, told employees at a recent company-wide meeting that Gemini would launch later this year.

In 1998, Sergey Brin and Larry Page, still Ph.D. students at Stanford University, were studying web search, and based on this research, they founded Google together.

The two have been working together since their Ph.D. studies at Stanford University

He is the second-largest individual shareholder in Google parent Alphabet after Larry Page, with a stake worth nearly $90 billion, according to S&P Capital IQ.

Since resigning, Sergey Brin has focused his attention on public goods (such as new forms of disaster relief) and airships.

Shares of parent company Alphabet have risen more than 10 percent since Google announced a slew of AI-infused products at its developer conference in May, such as a new version of its search engine.

This stock price increase is in line with the stock price of old rival Microsoft and the Nasdaq Composite Index.

After Sergey Brin and Larry Page quit their day jobs to promote Pichai to CEO of Google four years ago, they talked to him regularly, offering him advice and encouragement, but not nagging him every day.

Of course, they still control the majority of Google’s voting power and sit on the executive committee along with Pichai.

It is said that Pichai was very excited about the return of Sergey Brin and encouraged him to contribute more.

Google Brain and DeepMind combined

As early as 2011, Google Brain was established to study how to use deep learning to build widely applicable AI programs.

This year, Pichai decided to merge Google Brain and DeepMind. After the merger, DeepMind CEO Hassabis became the CEO of the new group Google DeepMind.

Many of Hassabis’ and Google’s DeepMind executives are based in London, making the London office the new agency’s de facto center of power.

Now, the return of Sergey Brin can be seen as a beneficial intervention in the transition period of Google’s AI team.

Many Google employees said that researchers welcomed Sergey Brin’s participation as a recognition of their work.

Gemini may become the strongest opponent of GPT-4

Previously, Hassabis broke the news to the media that Gemini merged AlphaGo and GPT-4-like large model Gemini, which will be released in a few months

One is an AI system that uses reinforcement learning to defeat human Go champions and create history, and the other is the most powerful multi-modal large model that currently dominates almost all large-scale model lists. Google wants to combine these two AIs.

To this end, Google DeepMind is ready to spend tens of millions of dollars, or even hundreds of millions.

After merging AlphaGo with the language capabilities of large models such as GPT-4, the problem-solving and planning capabilities of the Gemini system will be greatly enhanced.

And since Gemini integrates AlphaGO, which uses reinforcement learning and tree search, there will be many upgrades.

For example, reinforcement learning allows AI to solve challenging problems by learning from repeated trial and feedback; tree search methods help explore and remember possible moves in a scene, such as in a game scene.

In 2017, AlphaGo Zero quickly surpassed AlphaGo without using human data

In addition, other fields of technology (such as robotics and neuroscience) will also be integrated into Gemini.

Hassabis said, “It can be said 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.”

Gemini is said to have multimodal capabilities not found in previous models and is very efficient at integrating tools and APIs.

Moreover, Gemini will be available in a variety of sizes, designed to support future innovations in memory and planning.

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

At the Google Developers I/O conference last month, Google mentioned that from the very beginning, Gemini’s goal was multi-modal, efficient integration tools, and APIs.

At the time, Google’s announcement was: “Although it is still early days, we have already seen in Gemini a multi-modal capability that has never been seen in previous models, which is very impressive.”

For example, AI bigwigs like LeCun said that Transformer limits the ability of the language model too much to the scope of the text.

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

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

Clearly, Gemini is Google’s last stand.

References: