In the legend, Gemini, which combines the large models of AlphaGo and GPT-4, has accelerated the pace of application landing with the support of Blin. Will Alphabet-C launch a "counterattack" in the AI war?
Four years ago, the legendary figure of Alphabet-C, who "retreated to the mountains and forests," has been blown back by the wind of AI!
He is Sergei Brin, the co-founder of Alphabet-C and a legend in the technology industry.
Sergei Brin
On July 20th, The Wall Street Journal quoted internal sources from Alphabet-C, stating that Brin, who had already retired, has been visiting Alphabet-C's office in Mountain View, California, three to four days a week recently, working with researchers to develop the Alphabet-C AI model - Gemini.
Insiders said that Brin attended an internal AI conference at Alphabet-C at the end of last year, and now his participation frequency and intensity have further increased. He is currently working closely with a group of researchers to create the long-awaited Gemini:
Brin will discuss technical issues such as "loss curves" with Alphabet-C researchers, and he will also be involved in personnel matters, such as how to recruit sought-after researchers.
Perhaps due to Brin's involvement, The Wall Street Journal quoted insiders as saying that Demis Hassabis, the Alphabet-C executive in charge of the Gemini project, told employees at a recent company-wide meeting that the brand-new Gemini model will be available as early as this year.
Is the long-awaited Gemini, which combines AlphaGo and GPT-4 models, finally coming?
Intense Activity within Alphabet-C
On December 4, 2019, Brin officially announced his "retirement" and transferred control of the company to Alphabet-C CEO Sundar Pichai.
Perhaps Brin did not expect that just three years later, the popularity of ChatGPT would bring about an AI revolution, nor did he expect that this AI battle would force Alphabet-C to engage in "head-on confrontations" time and time again.
In February of this year, Microsoft announced the integration of ChatGPT into Bing and gained its own generated search experience under the promotion of GPT-4, which made Alphabet-C "shudder."
Many Alphabet-C employees have stated that since the AI boom began in November last year, Brin has frequently attended Alphabet-C's AI-related meetings. Brin's influence at Alphabet-C is continuously expanding, which indicates the intensity of the AI war. No tech giant is willing to "surrender," and it also shows Brin's enthusiasm for AI. It is worth mentioning that the well-known "Alphabet-C Brain" team originated from Google X, which was led by Brin in the early years. In a shareholder letter issued in 2018, Brin wrote, "The capabilities and potential of AI, as well as the potential to solve important problems through computation, have never been so enormous. The development of AI will be the most significant development I have seen in the field of computing in my lifetime."
According to the media, when Alphabet-C CEO "Chop" learned that Brin would join the Alphabet-C AI research field, he was very excited and encouraged him to make contributions.
Alphabet-C has always been regarded as a top artificial intelligence research center, but with the intensification of AI competition, this technology giant has entered a "defensive" mode - in order to resist a group of agile and increasingly large AI competitors, to protect its core search business, stock price, and potential future security threats.
According to Wall Street News, in April of this year, Alphabet-C's parent company Alphabet merged its two top AI teams, Alphabet-C Brain and DeepMind, and appointed the founder of DeepMind, Demis Hassabis, as CEO. At that time, Alphabet-C stated that this move would accelerate its AI progress.
Regarding the newly merged team, Haasabis is clearly very confident. He said, "The new team brings together two forces that are crucial to recent advances in artificial intelligence."
"If you look at our position in the field of AI, you would believe that 80% or 90% of the innovation in the future will come from one of these teams. Both teams have made extremely outstanding achievements in the past decade."
And this new product, Gemini, launched by the top AI team, can perhaps be called Alphabet-C's "last-ditch battle" in this AI war.
Rumored to be incredibly powerful, Gemini will combine AlphaGo and large language models, according to Alphabet-C DeepMind CEO Hassabis.
One is an AI system that defeated the human Go champion using reinforcement learning and made history, and the other is the strongest multimodal large language model that dominates almost all major model rankings. What kind of sparks will be generated when these two large models come together?
As previously mentioned by Wall Street News, Gemini will combine the language capabilities of AlphaGo and GPT-4, among other large models, and the system's problem-solving and planning abilities will be greatly enhanced.
Some AI experts believe that the main limitation of language models is their indirect learning through text.
However, AlphaGo's advantage can address this issue. In 2016, DeepMind's AI system AlphaGo defeated world Go champion Lee Sedol with a score of 4 to 1, becoming the first robot to defeat a Go world champion in history.
AlphaGo is based on DeepMind's pioneering reinforcement learning technology. This technology allows AlphaGo to repeatedly try and receive feedback on its performance, enabling it to handle complex problems that require selecting appropriate actions. Additionally, AlphaGo uses Monte Carlo tree search techniques to explore and memorize possible moves on the game board.
The next leap for language models may involve performing more tasks on computers. As mentioned in a previous article, Gemini's greatest advantage lies in its multimodal capabilities. It can not only understand and generate text and code but also comprehend and generate images. In contrast, ChatGPT is merely a text-based model that can only understand and generate text.
Furthermore, an important step in creating language models with capabilities similar to ChatGPT is to use human feedback reinforcement learning to improve their performance. DeepMind's extensive experience in reinforcement learning can provide Gemini with new capabilities.
During the Alphabet-C Developer I/O Conference in May, Alphabet-C mentioned that from the beginning, Gemini's goal has been multimodality and efficient integration tools and APIs. The preview at that time stated, "Although it is still in the early stages, we have seen unprecedented multimodal capabilities in Gemini, which left a deep impression."
Where will the next major leap for language models be? Gemini may provide guidance for the direction of the next generation of language models. Accelerating the launch of Gemini may increase Alphabet-C's chances of success in this AI competition.