"I'm not afraid to tell you that the DEEP-ID face recognition technology I have been working on for 15 months is just my test work in deep learning artificial intelligence, mainly to train the team. Pen Fun Pavilion wWw. ｂiquge。 infoFrom next year, that is, starting from 06, I will put the main direction of artificial intelligence on user behavior analysis and preference analysis.

Finally, the business model of 'using artificial intelligence to figure out what users like, so as to push their favorite content products to the other party'. I believe that when this day is realized, whether it is Dingdang.com, a certain point network, or NICONICO, the scale of genuine paying users will produce an explosion of several times or even dozens of times.

At that time, the content industry will usher in a 'pirated junk resources are easy to find, but pirated books, songs, and games that know your taste are hard to find, and users may not pay for pure content, but they have to pay for the artificial intelligence combination package of 'content + knowing what content he likes'. ”

Gu Cheng used a relaxed outlook to pull Ma Feng and Ding Sanshi into a leisurely and fascinating world.

Ding Sanshi is in the content industry, from games to music, and he knows how serious the piracy problem is in China. And if the problem of piracy can be solved, it will be enough to feed dozens or even hundreds of times more content industry players, and the long-tail effect of segmented demand will be brought into full play.

"You...... The vision of the future that you wrote in your book "The Long Tail Theory" turned out to be how you knew how to achieve it many years ago...... That's too much of an exaggeration, you didn't even hear the name of Professor Jeff Hinton when you wrote The Long Tail Theory, did you know it? There is no concept of 'deep learning artificial intelligence' in the world, how did you predict that there would be such a technology to realize your layout in the future?"

Ding Sanshi felt that his brain was no longer enough. He has never served a few people in his life, although he is very convinced of Gu Cheng, but until today, he found that compared with Gu Cheng, he himself is completely a frog at the bottom of the well.

Unless Gu Cheng is a prophet or a time-traveler, how could he have predicted two or three years ago that he would only see a little budding form of future media distribution today?

"You guessed it, before I wrote 'The Long Tail Theory,' I saw it today, and even three or five years from now. Gu Cheng admitted without modesty. He needed this kind of iron-clad prestige to make Ding Sanshi and Ma Feng believe in him.

"Then if you tell us all this today, aren't you afraid that we will also start to lay out and copy your routines like Ma Teng?" Ma Feng had a deeper understanding of human nature, and immediately felt that Gu Cheng seemed to have some fearless conspiracy.

Gu Cheng smiled heartlessly, and smiled very gentlemanly: "Copy, how to copy, big data is in my hands, do you have it? artificial intelligence algorithms are in my hands, do you have them? none of these, even if you know my thoughts, what do you copy?"

Ma Feng was stunned for a moment, and said in his heart that this is indeed the truth.

No wonder Gu Cheng was fearless.

Ma Feng secretly wondered that if Gu Cheng didn't have any purpose, he would rarely confide in his peers so much dry goods. Could it be that he had other intentions when he suddenly said this today? So he naturally asked:

"Then what is your plan to talk to us so much today? It can't suddenly open our eyes to literacy, right?"

Gu Cheng saw that he was exposed, so he no longer concealed it: "Brother Ma is a thief, you can see it at once." Okay, then I won't hide it from you, let's be honest. In order to operate well in the future, deep learning artificial intelligence will require a myriad of supporting conditions, and I will not be able to conquer the world by myself and do it from scratch to toe. So there are some supporting facilities, I think there will be a certain future in the future, but it is not worthy of me to do it personally, if you have the intention today, you can find resources to prepare by yourself. ”

Ma Feng and Ding Sanshi's ears stood up at once.

There are too many miracles created by Gu Cheng, even if he plucks the hair down, and he doesn't look down on the small business at all, there are unlimited business opportunities behind it.

Of course, the business that Gu Cheng doesn't look down on must be more difficult to do, or vulgar, or it can't form a monopoly and can't make huge profits all year round, and can only make quick money for a few years, and then it will fall into the fierce competition with a low threshold.

But even so, it is definitely a business opportunity that ordinary people can't ask for.

Seeing that his buddies were listening very carefully, Gu Cheng didn't hide it, but simply stretched out three fingers and gestured and said:

"In the future, to do a good job in deep learning artificial intelligence, there are three key conditions.

The first is to have a good algorithm, this is the core, just like software.

The second is to have sufficient big data, which is equivalent to the input material in the software. Moreover, deep learning software can be self-optimizing and self-evolving, the more input materials, the better the data labeling, the better and more accurate the final effect. Once big data in a certain aspect has an advantage, the future giants can quickly form a snowball effect and widen the gap with the latecomers.

Third, it is necessary to have the hardware processing speed to process these days of data - with the current amount of artificial intelligence computing, even if it is to solve the simplest picture recognition, or voice intelligent recognition, or even just input method frequency statistical feedback, it is not something that any single server can carry. Perhaps every demand has to be handled with the processing power of a data center.

However, at the same time, there may be a high repetition rate of multiple problems processed by each data center, and the reuse rate of data processing process volume will be high. Therefore, enterprises with more processing needs and more processing hardware resources will have obvious advantages in processing efficiency. At this time, we will find that relying on the traditional 'buy a batch of servers for each new business need and build a dedicated data center' will be very inefficient and very costly......"

As soon as Gu Cheng said this, he was interrupted by Ma Feng, who didn't know much about technology: "So how to solve this problem? If it's just expensive money, it's not a big problem, we spend money, others also spend money, can't we just raise the price together?"

"Huh...... If it's 'you're expensive and I'm expensive', it's really not a big problem, just raise the price. However, if the application cost of artificial intelligence is too high, its substitution effect on traditional industries and traditional consulting manual jobs will not be obvious. The cost of using artificial intelligence to solve problems on a large scale must be significantly lower than that of hiring human customer service, reporters, editors, teachers, consultants, grassroots lawyers who check legal documents, and grassroots accountants who can only issue invoices......

Only then will the capitalists dismiss these grassroots white-collar workers who repetitive mental work and use robots instead. If you don't solve the cost problem, you may not be left behind by other AI competitors, but the process of eliminating human jobs in the entire AI industry will cause at least three to five years of delay. ”

Ma Feng was still pondering the calculation, but Ding Sanshi had already understood: "Indeed, just like this year, I saw Binjiang next door to our Huangyi headquarters, there are several companies that do LED in emerging industries, although everyone is expensive, but the main reason why they can't be sold is not because it is more expensive than their peers, but 'after buying LED products, the price minus the electricity cost saved in the whole life cycle of the product is still more expensive than energy-saving lamps', so those property owners don't buy it." I've had a few kung fu teas with the owner of one of the companies called Roledis, and I've heard him talk about it.

There is also Hikvision next door, which was also the case in the early years - a set of monitoring is so expensive, and the reason why it can't be sold is not because it is more expensive than its competitor Dahuajia, but because the cost of installing these more sets of monitoring is more expensive than hiring one more security guard, so it can't be sold. In the past two years, the cost has come down, and those savvy business owners have begun to install more surveillance and hire less security guards. ”

Ma Feng's knowledge is still very extensive, he started as an e-commerce company, and he knows a little bit about all walks of life. After listening to the example given by Ding Sanshi, he immediately figured it out.

The popularization of artificial intelligence is, in the final analysis, the price of PK with labor costs, not the price of PK with peers. Only when it is cheaper than people can it be used by capitalists.

So he asked Gu Cheng sincerely: "Then Xiao Gu, do you think there is any way to quickly reduce the cost of artificial intelligence computing? We can't let us invest in Intel and expect hardware manufacturers to accelerate and drastically drop the CPU price per unit of computing efficiency, right?"

"Of course not. Gu Cheng picked up the teacup and took a sip with a calm expression, "First of all, the architecture mode of the CPU, when processing the parallel data processing required by deep learning algorithms, the efficiency is still too low. Even if you want to start with hardware, you shouldn't look to Intel. ”

You should find Nvidia, but Gu Cheng cherishes himself with this sentence, and there is no need to tell Ma Feng and Ding Sanshi.

He acquired AGIEA in 02 and has taken the lead in the layout of GPUs for artificial intelligence. In the future, whether it is digging NVIDIA's technical team, accumulating technical reserves, or directly finding a way to invest in NVIDIA, it is very flexible.

He can do this by himself, and he doesn't need the cooperation of Ma Feng and Ding Sanshi.

What he needs to say today are some hard bones that he doesn't want to gnaw.

"Secondly, even if we don't strive for leapfrog development in the physical indicators of hardware, we can still make a fuss about the efficiency of resource allocation - this solution is called cloud computing. Take advantage of the further convenience of the future network to integrate the data centers of major Internet companies.

Or let a specialized server/data hosting company form a virtual server/virtual data center architecture, optimize the scheduling of computing resources in real time, and temporarily and time-share lease idle computing power to artificial intelligence companies that have processing needs at a relatively low price - with these tricks, it is easy to save several times the cost of computing resources by optimizing the allocation without a leap in the physical performance of the hardware. ”

"The trick you said can be technically implemented so far?" Ma Feng and Ding Sanshi said in unison.

Gu Cheng shrugged calmly: "It's more difficult, but there is already a chance - at the beginning of the year, Intel began to push dual-core CPU processors, and the allocation efficiency and architecture of multi-threaded computing have also been greatly optimized." Based on this basis, our company has experimented with a 'distributed compilation' architecture of an internal code compilation server. If you don't believe me, you can take a visit. ”