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

First, actively use AI to solve various problems, including translation, maps, language, images, etc., and incorporate AI into existing Google products and services to enhance their capabilities and user experience.

Second, continuously release various AI software and related tools, make AI technology easy to use and accessible, continuously share the latest research results, publish papers, and open-source code. Let more researchers benefit and allow developers to better develop their own machine learning models using the AI tools developed by Google.

Third, use AI to solve major problems faced by the world, such as healthcare, energy, environmental protection, etc., and collaborate with other organizations to jointly address challenges.

Since AI has not been widely implemented in the industry yet, Jeff's plans and routes seem a bit empty.

After listening to Meng Fanqi, Jeff was stunned by his more realistic and specific plans.

"I plan to quickly launch AI models and technologies to solve various practical problems. The long-term goal is to solve them in a universal and versatile way. For example, in the field of language, translation, question answering, classification, generation, one model can handle them all.

Not only should the language model be versatile, but the visual and speech models should also be ultimate giants, directly integrating classification, detection, segmentation, and generation.

First, create a visual storm, revolutionize image tasks, then unify the paradigm of language models, and integrate multimodal capabilities, combining language, images, and sound."

Meng Fanqi spoke confidently, leaving Hinton, who was still contemplating how to design better neural networks, and Jeff, who was leading a large team for the first time, dumbfounded.

Jeff was dizzy from listening. Wow, your plans are even bigger than mine? This little guy is blowing with such confidence.

When I return to Silicon Valley in a few days, I'll blow this rhythm with Google's CEO!

Revolutionize machine vision in one year, unify the language paradigm in two years, integrate multimodal capabilities in three years, and in the fourth year, a large model will dominate the industry.

Will we achieve fully autonomous driving in the fifth year, completely intelligent robots in the sixth year, and create Terminators in 2020, the seventh year?

In the tenth year, we'll directly invade virtual alien spaceships.

Find the original at "pawread dot com".

Jeff never expected that Meng Fanqi wasn't just bluffing.

Although he exaggerated a bit in terms of time, such as the ChatGPT model with billions of parameters, it would be difficult to train without Nvidia's A100.

The more advanced and powerful the GPUs, the more cards are needed for distributed training. The more cards there are, the more likely hardware issues will arise. If a few cards encounter errors, the entire process may be affected.

Based on Nvidia's release schedule, it is likely that ChatGPT, a conversational AI at this level, will be limited by hardware and may not be available until the end of 2020, at most two years earlier.

However, in order to once again advance the release of this series of technologies, Meng Fanqi could consider sacrificing some of ChatGPT's performance, thereby bringing forward its release by one or two years.

Starting from now, four years may have been a bit exaggerated, but the prospects in the fifth and sixth years are still very promising.

But Jeff is an honest person. He knows very well that he has been drawing empty promises.

Initially, the Google Brain team did not have a high status. Although Andrew Ng founded the project and demonstrated the potential of deep learning, the team did not have much output for a considerable period of time.

Many Google executives believed that it was a waste of money and held skeptical and negative attitudes towards the team's direction.

In 2012, the breakthrough achievements of Hinton's team and Hinton's joining at the beginning of the year greatly alleviated this problem, and everyone still held great respect for this AI godfather.

However, although Hinton improved efficiency and quality through his oversight, there was still no phased achievement in the implementation of technology. No practical projects truly utilized the new technology.

By absorbing Meng Fanqi this time, in addition to leveraging his amazing results to elevate the status of the Google Brain team, Jeff was very interested in the real-time detection algorithm they just discussed.

This could be a large-scale technological application that could completely change the Google Brain team's trend of only making small waves.

"I'm sure Baidu's CEO, Li, has already summarized the situation in all aspects regarding the real-time detection algorithm." After discovering that Jeff was also interested in this matter, Meng Fanqi laughed heartily. The technology he carefully selected for monetization and deployment at the current stage was indeed attractive.

"Li's PPT and technology conference must have been very well-prepared. His presentation is much more detailed than what I can say directly. Let's watch Baidu's technology conference on Friday."

What a joke, we haven't even seen the contract for the revenue split yet, and they want me to reveal the technology? Meng Fanqi remained calm and started playing Tai Chi.

Although it was already decided that he would sign with Google due to the need for Google's framework and TPUs, it was impossible to discuss technical details before signing the contract.

If the revenue split agreement with Google turns out to be really unsatisfactory, Meng Fanqi can completely withhold many technical contents. Even with contractual restrictions, it wouldn't be so difficult to bypass them and publish through friends.

While Google was just starting to show interest in the real-time detection algorithm, making an early move, Baidu, which had reached an ideal stage of negotiations with several leading companies, was making rapid progress.Due to the technical nature of this service, the actual needs of the clients vary greatly. It often requires on-site operation, debugging, and demonstration at the client's location, which most Baidu employees cannot handle in a short period of time.

Instead, it's the Baidu Deep Learning Research Institute, originally a research-oriented department, that has mobilized all its staff. The researchers have turned into sales and customer service representatives. This is a microcosm of domestic tech giants' eagerness to monetize their technology.

Up to now, the Ali and Penguin groups have not expressed any specific needs, but they have begun to realize the importance of AI technology and hope for technical exchanges. The two sides have already entered the third phase of discussions.

As for the large manufacturing and coal and metal industries, they still need to provide some data for model training. However, the current negotiation situation is quite good, with high expectations on both sides.

The most crucial part is the demand related to road monitoring and security from the government. If this order is secured, it would be a significant achievement.

Government orders are stable and long-lasting, and they generally do not change frequently. Especially in this area, the order volume is much larger than that of enterprises, and the requirements are usually lower. There might be more paperwork and minor issues, but they are entirely acceptable.

The team responsible for liaising with the government is led by Vice President Yu Kai, and it includes a group of elite candidates.

Among them is Wang Kai, who has directly liaised with Meng Fanqi and consulted him on technical issues multiple times.

At this moment, he is sitting in the Public Security Bureau, looking around anxiously.