Sensing Meng Fanqi's unique insights, Jeff also shared some of Google's current internal AI concepts and goals:

First, actively use AI to solve various problems, including translation, maps, languages, images, etc., and try to integrate AI into existing Google products and services, and use AI technology to improve product capabilities and user experience.

Second, we will continue to launch various AI software and related tools to make AI technology easy to use and accessible, continue to share the latest research results, and continue to publish papers to the public and open source code. Benefit more researchers and allow developers to use the AI tools built by Google to better develop their own machine learning models.

Third, we will use AI to solve the big problems facing the world, such as healthcare, energy, and environmental protection, and cooperate with other organizations to solve the challenges together.

Since there are not many contents that AI can really land in the industry at present, Jeff's pies and routes are a bit fake.

After Meng Fanqi listened, he directly backhanded and began to draw cakes for Jeff, and the painting was more realistic and specific.

"I plan to quickly roll out AI models and technologies to solve all kinds of practical problems, and the long-term goal is to solve them in a universal way. For example, language classes, what translation, Q&A, classification, generation, a model does it all for him.

Not only should the language model be a universal model, but the visual speech model should also be an ultimate giant, and directly generate a four-in-one for him to directly classify and detect what to detect.

First, it sets off a visual storm, revolutionizes the image task, and then unifies the paradigm of the language model, and then integrates multimodality, and the language picture and sound are directly integrated. ”

Meng Fanqi talked eloquently, and Hinton, who was still thinking about how to design a better backbone neural network during this time period, and Jeff, who led such a large part for the first time, were stunned.

Jeff was dizzy when he heard this, good guy, you have a lot more pie than me, huh? The little things blow in a good way.

When I go back to Silicon Valley in a few days, I'm going to blow to Google's CEO at this pace!

One year of revolution in machine vision, two years of unified language paradigm, three years of multimodal integration, and the fourth year of direct domination of a large model.

Is it the fifth year to directly complete the full level of autonomous driving, the sixth year to be fully intelligent robots, and the seventh year, 2020 to create a terminator?

The 10th anniversary of the direct virtual invasion of alien spacecraft.

Jeff never thought that Meng Fanqi was really not painting cakes.

Although the time is a little braggart, exaggerated. For example, a large model with 100 billion parameters at the level of ChatGPT will be more difficult to train without NVIDIA's A100.

If the cards are not good enough and not advanced enough, the more cards are needed for distribution. The higher the number of cards, the more likely there is a hardware problem, and once a few cards have an error, the entire process can be affected.

Judging from the release time of NVIDIA, a conversational robot of the level of ChatGPT is likely to be limited by hardware and will not come out until the end of 2020, at most about two years in advance.

However, in order to launch this series of technologies in advance again, Meng Fanqi can also consider slightly sacrificing some of ChatGPT's performance, so as to bring forward the emergence of ChatGPT by another year or two.

From now on, four years may still be a bit of a boast, but five or six years is still a lot of hope.

But Jeff is also a sincere person, and he knows that he is suspected of painting cakes.

The Google Brain team didn't start out as a high-profile team, and while Ng founded the project to demonstrate the potential of deep learning, the team didn't produce much for quite some time.

Many high-level people at Google think it's a waste of money and are skeptical and negative about the direction of the team.

The breakthrough results of the Hinton team in 2012, as well as the addition of the Hinton team at the beginning of the year, have greatly alleviated this problem, and everyone still has great respect for this old AI godfather.

However, although there is a Hinton check on the research, the efficiency and quality have improved a lot.

There has never been a phased achievement in the major event of technology implementation, and there are no actual projects that really use new technologies.

In addition to once again improving the status of Google Brain's team with his amazing results, Jeff was very interested in the real-time detection algorithm just now.

This is likely to be a large-scale technology application that can completely sweep away the decline of the Google Brain team that has been only making small fuss about.

"In the matter of real-time detection algorithms, Mr. Li of whiteness must have summarized the situation in all aspects in great detail." After discovering that Jeff was also playing with the idea of this matter, Meng Fanqi laughed, the technology he had carefully selected for the fastest monetization and deployment at this stage was really attractive.

"Mr. Li's PPT and technical conference, presumably very thoroughly, his display is much more detailed than what I said directly, let's just watch the technical conference of whiteness on Friday."

What are you kidding, you haven't seen the contract yet, and you want me to pay for the technology? Meng Fanqi remained silent and started Tai Chi.

Although it is a foregone conclusion to sign Google early due to the need for Google's framework and TPU, it is impossible to talk about the technical details before signing the contract.

If Google's sharing agreement is really unsatisfactory in the end, Meng Fanqi can completely avoid handing over a lot of technical content, and this kind of purely technical and ideological stuff, even if there are contract provisions, it is not so difficult to bypass the restrictions and publish it through friends.

Google here has just moved its heart to the real-time detection algorithm and attacked in advance, and Baidu, which has reached technical cooperation with Meng Fanqi, has already discussed a more ideal stage with a number of leading companies, and the progress is very fast.

Due to the strong technical nature of the service, the actual needs of customers are many different from each other, and it is often necessary to go to the customer's side to operate, debug and demonstrate on site according to the requirements, and most of the employees of Baidu are not qualified for the job in a short time.

It's the Baidu Deep Learning Research Institute, which was originally part of the nature of the research institute, and the researchers directly became sales + customer service. This is also a microcosm of the eagerness of domestic manufacturers to realize technology.

So far, the Ahri and Penguin lines do not have special needs, but they are beginning to realize the importance of AI technology, so they hope to have a technical exchange, and the two sides have talked about the third stage.

Several large manufacturing and coal metal industries still need to provide some more data and do some model training. However, the current negotiations are quite good, and expectations on both sides are high.

The most important is the government's road monitoring and security-related needs. If this part of the order is won, it will be a big deal.

Government orders are stable and long-lasting, and are generally not likely to be replaced frequently, especially in this area, the order volume is much larger than that of enterprises, and the demand is usually lower than that of enterprises. There may be more formalities and trivialities, but it's perfectly acceptable.

The relevant parts of the government are responsible for docking, and a group of elite candidates personally led by Vice President Yu Kai.

Among them, it naturally includes Wang Kai, who has been directly connected with Meng Fanqi and has contacted many times to consult technology.

At this moment, he is sitting in the public security bureau, looking left and right.