The meeting lasted about forty minutes, and Chen Yu assigned tasks to the major development teams.

There are also many project teams in the company's technical department, involving natural language processing, recommender systems, image generation, deep reinforcement learning, computer vision, and robotics.

Quantitative capital, the outside world has identified the company as an investment management company, but it is actually a high-tech company disguised as an investment management company.

As the meeting ended, everyone else left the conference room, leaving only Fang Hong and Chen Yu.

"I don't think there's any problem with the development route you've set." Fang Hong said with a smile: "But it can't be slowed down, I have reliable information here, Google's side, their artificial intelligence lab GCP put forward an idea at an internal meeting, that is, to make machines 'learn' language and grammar, and they are also working on language models just like you." ”

Chen Yu was surprised: "Google is also doing it? What is the level of technology? ”

As the real behind-the-scenes boss of Qunxing Capital, Chen Yu believes that he should have reliable information when he says this.

Fang Hong said calmly: "It's not as advanced as you, I can be sure of this." ”

According to the memory of a past life, around October this year, Google should release a GPT-based language model, which learns language through the encoding of input, mainly a pre-trained language model, but Google has not developed such a language model before.

Hearing Fang Hong say this, Chen Yu was immediately relieved, this field is one step ahead and often step by step, and the development speed of AI is often developing rapidly at an exponential growth rate.

After a while, Chen Yu looked at Fang Hong and said: "Our AI model, the first generation of iterative research and development task is to achieve the ability to generate human text, supported by tens of billions of parameter scale, and has now entered the second generation of research and development, which is larger and stronger than the first generation, and can produce coherent and attractive responses to various prompts, with a parameter scale of 50 billion." ”

"As for the third-generation model, which should need to be supported by 200 billion parameters, the goal of the third-generation model is to be bigger, more powerful, and more advanced, and achieve amazing performance on multiple natural language processing tasks."

"For the upgrade of AI large models, it needs the support of strong underlying capabilities to play, and to break through the bottleneck of AI through large computing power and large-scale training data, the operating costs are also very huge, and now the operating costs in one month exceed 300 million yuan."

Hearing this, Fang Hongyan said succinctly: "Money is not a problem, it's not enough for me to manage." ”

Chen Yu smiled: "It's really not enough, I'll look for you." ”

Its underlying capabilities include computing resources, data storage, power supply, and network bandwidth.

These are very money-burning things.

AI training requires the use of large-scale computing resources, including GPUs and GPUs, and cloud computing can provide them with powerful and elastically scalable computing resources.

The data center at the bottom of cloud computing can provide high-performance computers and hardware equipment, and can be optimized to improve the utilization rate of computing resources, thereby accelerating the training and generation speed of AI, and at the same time, it can also provide powerful parallel computing capabilities, and multiple computing tasks can be distributed in different computing nodes at the same time for calculation, improving training efficiency and speed.

In addition, training AI requires network requirements, one is to achieve perfect link load balancing, and the other is to be able to recover quickly in the event of a network failure.

Because of the "barrel effect", as long as one link has an uneven load and causes network congestion, which becomes the shortcoming of the barrel, then even if the other two channels are unimpeded, the collective communication time will still increase significantly, thus affecting the training efficiency.

It is similar to the difficulty of one core and the onlookers of many cores.

As the size of the cluster increases, link failures are inevitable, and a similar link failure will cause the entire communication to stall, which in turn will cause the model training to stagnate, so it is necessary to quickly repair the failure.

Fang Hong leaned back on the chair with Erlang's legs, took advantage of the armrest of the chair with one hand, and said thoughtfully: "This AI of yours is still one element away." ”

Hearing this, Chen Yu was curious: "What's the difference?" ”

Fang Hong said succinctly: "Image." ”

Chen Yu was slightly stunned: "Image? ”

Fang Hong immediately began to say: "How to make the product more quickly accepted by the public? How to form user stickiness, especially to capture the younger generation, you have to make this AI visible, not a string of 0 and 1, so you should give AI a character image, that is, anthropomorphize it, to capture the younger generation, especially the post-90s and post-00s generation, and further say that AI should be given a two-dimensional character image. ”

Hearing this, Chen Yu couldn't help but nodded: "It makes sense." ”

After a while, he looked at Fang Hong and said: "I know how to do it, since I want to do it, let's make it big, and I will set up an additional department to create a virtual image brand operation department for AI, create an exclusive IP image, enrich its personality, and have a complete story background." ”

"Including inviting some professional animation production teams and screenwriters to produce some excellent story fragments to enrich their character images, and the publicity can be crowned as the first aboriginal of the 'virtual world' and so on."

Fang Hong said: "It is not easy to create an IP from scratch until it is accepted by the public and deeply rooted in the hearts of the people. ”

Chen Yu nodded, and when Fang Hong pulled it so little, he immediately understood the importance of AI visualization and anthropomorphism.

And it is promising, and the future launch of APP, which can become a must-install virtual assistant for users, including smart home Internet of Everything, and various home appliances and smart devices in the future can be hosted to virtual AI assistants.

Now to set up this IP image and with the growth of the younger generation, there is a lot of room for imagination in smart home appliances alone, air conditioners, washing machines, computers, mobile phones and other equipment, these manufacturers are not capable of AI, and only need to access the AI assistant of quantitative capital when the time comes.

In the future, if it cannot support access to this virtual AI assistant, users will not even want to use products such as washing machines or air conditioners of this brand.

After all, there are many brands of washing machines and air conditioners, but there is only one such a powerful AI assistant, and the IP avatar created has grown with a generation and formed user habits.

At that time, those manufacturers will not be able to access it, and it is impossible for quantitative capital to make air conditioners and washing machines by themselves, which is not a bad thing for manufacturers, and they can save themselves from being forced to enter unfamiliar fields in order to build the so-called moat to invest heavily in the construction of AI and ecosystems, and they only need to do their own products well and finely.

……

Fang Hong stayed in Quantitative Capital for an hour and a half before leaving, and these tracks were handed over to Chen Yu, who was very confident in him.

Back at the Meditation Villa, Fang Hong went to the study, presented the fame system data panel in front of him, and then directly consumed 100,000 achievement points to exchange for a prop card to specify Chen Yu's character attributes and check it.

[Target: Chen Yu]

[Character: Excellent]

[Prestige: 85105209]

[Health: Excellent]

[IQ: 145, genius]

[Coordination: Excellent]

[Wealth: 17.6 billion RMB]

……

"This hexagonal data, not bad." Fang Hong was very satisfied when he saw the data of the six major attributes, especially the IQ value reached 145, as long as the value above 140 was a genius level.

The most important thing is that Chen Yu's data are all initial data, and they have not been strengthened and improved by Fang Hong.

Especially the IQ level, between 120 and 139 is an excellent level, and the level of most senior talents is around 130 to 139, and those who have been promoted by Fang Hong to the level of leading talents have an IQ level of over 140.

Chen Yu's initial data level surpassed those who were promoted to the leading level, which is very rare.

In the end, Fang Hong decided to raise Chen Yu's IQ level to the 150 level, and the further he increased by 1 unit, the greater the consumption of the corresponding prop cards.

Chen Yu's initial data was so good, which indirectly saved Fang Hong a lot of achievement points.

After a while, Fang Hong exchanged a large number of R&D item cards, all of which were designated to improve Chen Yu's R&D ability, and the data showed the improvement of his IQ level.

After raising Chen Yu's IQ level to 150 before stopping, Fang Hong glanced at the achievement points consumed and couldn't help muttering: "Actually consumed 12.6 million achievement points......

It takes 100,000 achievement points to exchange for an item card, and it increases Chen Yu's IQ level from 145 to 150, which increases by 5 units, but consumes 126 item cards.

However, Fang Hong is not distressed, the more famous Chen Yu will be in the future, the greater his prestige and social influence, and the greater his prestige and social influence will also indirectly accumulate fame and achievement points for Fang Hong, and the more than 12 million achievement points consumed here will definitely return to the cost, and even earn a lot.

……

(End of chapter)