Chapter 258: On the way to serve food, you can still cook while walking
"Su Fat's technical route is actually like this: he promoted the 'artificial intelligence to recognize video content and push it according to user preferences' mentioned in the PPT, but in fact it was not done at all.
Jack, I don't know if you've noticed it - why do the videos on the fast hooves have so many words on the homepage? such as this 'teach you how to tie the ball head', or the 'braces girl, how beautiful is it', 'Taimu human~'. From an aesthetic point of view, don't you think it's ugly to type so many words in a short video?"
Feng Jianxiong induced eloquently, trying to let Jackma find out for himself.
"yes, why, it's ugly. Jackma went along the lines of thought.
Feng Jianxiong smiled slightly: "Because in fact, Su Pang is the same as Zhang Yiming, he can only recognize the title and text description of the video." His artificial intelligence can't recognize the story content of video images. He just read the text and then thought the video was about something, so he recommended it to other users who 'used to like to read similar text titles or descriptions'.
Therefore, the truth of Su Pang is just to follow Zhang Yiming's technical route, and then disguise it as being able to intelligently push videos. At present, there is no artificial intelligence on the earth that can understand the storytelling of videos. Not for at least four years. ”
"I see...... Indeed, at present, the strongest in the field of image recognition in the world is Google, followed by our Ahri department, I said how do they make the machine understand the video content? ”
This sentence was not said by Jack Ma, but by Lao Zeng in order to remedy it.
He seems to have forgotten that not long ago, he swore that "Su Fat can also do it, so you Feng Jianxiong don't deserve a high price".
No one cares about that, though.
Jackma is not stupid either, he knows that his subordinates are there to help him bargain.
No boss would punish a subordinate for lying to help him bargain.
Feng Jianxiong is also a top person, he smiled at the moment, and he couldn't chase after him.
He also gave the sweet dates after drinking the first stick together: "Lao Zeng finally figured it out - Lao Ma, you just think about it, the reason why Google does a good job in image content recognition is because they want to engage in an intelligent image search engine."
The reason why you Ahri are so invested is because you want to let the woman go to the offline clothing store to take a photo, and then go to Taobao to find the same style.
At present, there are not many commercial monetization channels for these technologies. So it's just you and Google throwing money at it. Even you haven't even smashed out the results of a single hair, and Su Fat's little body can make a fart. Scientific research is real money, and it is not a trick. ”
This sycophant is very subtle, and the key is to have something to say, which makes Jack Ma listen very well.
After all, Feng Jianxiong praised his strategic vision.
For a forced king, the general sycophant can't listen to it.
But if someone can demonstrate the wisdom and foresight of his strategic decisions from the perspective of high-tech development trends, then Jack Ma will be so cool that his pores are transparent.
It is a pity that the world no longer has such eloquence and sharp vision.
"Hmm...... It seems to be reasonable," Jackma pondered, and then thought of another question, "Xiao Feng, then how are you going to solve the problem of 'the current artificial intelligence can't read videos'? If you don't have a unique secret, it's just that Su Fat can't do it, and you can't do it." ”
"I can't do it right now, but I can do it in two or three years, and make sure it's at least one or two years faster than them. Feng Jianxiong Zhizhu said in his hand.
"Then describe your technology blueprint. Jack Ma asked in a nutshell.
Feng Jianxiong was also polite: "Yes, but can you let them avoid it first? My plan is not solved purely by technical means." It's about the lack of technology, business and operations. So, there are a lot of ideas involved that are easy to 'borrow'. I have to make sure of my trade secrets. ”
When Jack Ma heard this, he gave Lao Zeng and Cai Chongxin a color and asked them to go out for cocktails first.
Cai Chongxin and Lao Zeng glanced at each other, and left without saying a word.
Feng Jianxiong smiled slightly: "Actually, my method is simple to put it bluntly - as far as I know, the expected project period of Google's 'intelligent understanding video' project will take 5 years to overcome all the technical difficulties."
However, the past five years have not been '100 miles and half 90', but have been conquered separately according to each technical stage. The ultimate goal is to use artificial intelligence algorithms to identify which frames in a video are the 'main content/main idea' that determines the tone of the whole video and the theme of the story.'
Then, through careful reading, we can identify how many characters, what actions, and what stories are happening on the screen, and finally conclude that "this video is roughly telling a story". ”
When Feng Jianxiong said this, he paused slightly. The acceptance of Jack Ma was also observed, and several points were explained in layman's terms.
His theories, if they were all expressed in terms, would have been rather obscure.
However, to take an example that all primary school students understand, a horizontal comparison makes it popular - Google scientists, when they are training "deep learning" artificial intelligence and understanding human language/image information, are actually a bit like teachers giving Chinese lessons to primary school students.
I believe that primary school students remember that when they were in Chinese class, the teacher would take the trouble to ask you: What is the main content of this text, what is the central idea, and what are the clues?
Many elementary school students must have had 10,000 alpacas galloping by in their hearts: Nima! Lao Tzu knows what this text says, so wouldn't it be nice to understand it? You ask about the "main content" and "central idea" of a fart! If you answer the test incorrectly, you will deduct points!
But don't say it, when teaching robots to learn Chinese, you really have to strictly follow the "main content" and "central idea" step by step summary. Many humans feel that they can bypass or omit steps by instinct, but machines can't bypass.
Jackma quickly understood, motioning for him to continue: "So, then what?"
Feng Jianxiong continued: "Now that we know Google's technical roadmap, we can infer that before achieving the ultimate goal, Google needs to solve the problem of 'reading the story in a picture', and then consider 'how to choose the frames that best reflect the central idea in a video'.
So, how far is Google from the step of 'reading the story in a picture'? I think it's about three years - Google can now recognize whether there is a face in a picture, but it can't identify who this person is or whether the person in another photo is the same.
Google can also identify 'whether there is a cat in the picture', but it can't identify 'whether this cat is the same as the cat just now'. And after about three years, none of this is a problem. Google can do it 'from recognising that there's a cat in the picture' to 'recognizing what story this frame is telling'.
If we keep up with Google's pace and combine the image recognition research and development of Ahri, we can leave more than a year to spare. Then in 2016, we will also be able to come up with a commercial technology that 'allows the machine to read a frame'. Of course, this definitely requires in-depth cooperation with the Ahri image recognition team, as well as follow-up investment.
For me, as long as I can 'recognize one frame', I can 'roughly read the whole story', so that the problems that Google needs to solve by purely technical means can be temporarily bypassed by operational means. ”
If the technology is not enough, the operation will make up.
Feng Jianxiong has tried and tested this trick many times.
"How to make it up?" Jack Ma's tone was already a little eager, and he felt that he was completely impressed by Feng Jianxiong's genius brain.
"Jack, have you watched the video of Station N? Have you seen Youku or youtube?" Feng Jianxiong did not announce the answer, but asked rhetorically first.
"I've seen them all, not much. Jackma didn't know why, so he replied more cautiously.
Feng Jianxiong finally came to the moment to uncover the mystery, and he smiled proudly: "Then I don't know if you have found a detail - the videos on Youku and youtube actually do not have a 'cover', and the UP master of our N station, when uploading the video, has developed the habit of sending another picture as a cover - of course, to be honest, this habit is not spontaneously developed by the UP masters, but I asked Tianyin to force the UP masters to develop it."
Maybe in the future, when someone sees this trick working, they will follow me. However, it is impossible to improve the content provider level of a website overnight, and user habits cannot be changed overnight. Even if Youku learns it in the future, their UP master's level in choosing covers is far weaker than that of Station N......"
People who have watched the video of Station B in later generations, if they are also Youku users, I am afraid it is not difficult to find a phenomenon:
On station B, there is a group of bad people who often take covers to achieve the purpose of "title party".
For example, all MMD cloth models of Hatsune/Patrol/Weak Tone/Ai-chan dance animations will be covered with a particularly shameful picture of exposure, and users will be tricked into clicking on it first.
And then in fact, the scale of this picture is obviously beyond the scale of the video review, which belongs to "if there is such a lens in the video, it will definitely be 404".
At this time, users will shout angrily, "Surrender and don't kill!", "Shameful playback!", "The playback of the next work!"
The UP masters of station B who sang the sweetness, in order to cheat clicks, have since gone farther and farther on the road of the "handing over the seal and not killing" title party.
However, this is actually an operational conspiracy of station B...... Well, or to put it nicely, strategy.
Because it successfully forced its UP masters to carefully select a picture that best highlights the theme of the video to be used as the cover.
In contrast, a few years later, Youku videos, although they also began to have the option of "cover", instead of directly taking "the first frame image of the non-black screen directly as the default cover". But the cover on Youku is not so carefully selected and titled.
This differentiation strategy of station B is actually a bet that when the artificial intelligence picture reading technology develops to the threshold of "being able to understand the theme story of a certain picture", but it has not yet reached the level of "identifying which frames are most important in a video", during this window period, station B can be in the field of "accurately guessing the user's preferences for video content", to achieve sneaking.
What is the principle of stealing?
It's just that the machine doesn't know how to choose which frame in the video is the most important and the most relevant to the subject. But humans know, and the author of the uploaded video knows. Bilibili forced the author to complete the work of "a frame of selected representatives", and then the artificial intelligence algorithm was only responsible for concentrating on interpreting and classifying this frame.
Perfect realization of "in the case that the technology is not as good as Google, rely on operation to catch up with Google".
Feng Jianxiong is now planning for Station N, which is this method.
Brother knows that artificial intelligence cannot be achieved overnight, but Brother knows how to maximize the value of semi-finished products.
Google's technology is indeed better than him, but Google is too obsessed with technology and doesn't know how to "cook before the board".
You can continue cooking on the way out of the kitchen, even after the food is on the table.
And no one says that all dishes must be 100% finished before they can leave the kitchen, right?
Brother is not as fast as you to cook, but brother plays Japanese food, and you play French food, so brother can still serve food faster than you.
By extension, in fact, from this point of view, although the "shock department" of the UC department is shameless, from the perspective of developing artificial intelligence, the existence of the shock department is the same as the "handover and not killing" title party, which is a petri dish of "algorithm confrontation".
They can unearth the impulses of human nature in advance, and let the technical nerds test and plug the leaks.
Jack Ma stared at the dog leash.
He said incredulously, "This ...... So, in the future, if Station N is really hacked into a 'handover but not killed' website, how can you do it?"
"What's that, if it's really time for the algorithm to be tuned, I'll unload the mill and kill the donkey - directly say that I want to rectify the title party, and for the content uploaded by the UP masters that has nothing to do with the cover picture, the account will be banned for a certain period of time, or the list/advertising resources will not be given as a warning. It will be easy to govern when the time comes.
Actually, I know you're planning to buy UC Technology as a wholly owned company this year. If you have a good time working with me, I will not hesitate to teach you how to put those operations of tuning artificial intelligence algorithms on the side......"
For example, how to adjust the shock department, when to catch it, and when to put it. When do you sincerely apologize to society, and when do you secretly treat UC Consulting readers not as human beings, but as guinea pigs......
The use of wonder, in one mind.
"Heh, hehe...... That's it, I know why I haven't said I've ever won you in my life. Jackma felt that he was brainwashed as if he was in the clouds, and he didn't know what to do, "You said that Station N is currently worth $6 billion before partial financing, so it's 6 billion." I will invest another 2 billion and account for 25% of your shares. However, I still hope that you can cooperate with us in the field of 'public opinion distribution artificial intelligence' all year round, and I will not treat you badly. ”
"Brother Ma is bright! $2 billion is now smashed, and of course I won't let you down. Let's go and have a drink. Feng Jianxiong patted Jack Ma on the shoulder very casually, and left the lounge with him on his shoulders and laughing.
In fact, it is possible that Jackma chose to "copy Feng Jianxiong's idea, and then support other peers to go to the cottage".
However, the Ahri department does not have a "son" level layout in the field of entertainment. If this trade secret is leaked, it will be used by the two families, and it will definitely not be as valuable as it is now.
Jackma is a big person, and he certainly knows that investing is in people with potential, not companies and businesses.
Feng Jianxiong's potential is worthy of him "listening to the other party's creativity, and he will directly give money".
Zhou Tianyin followed the two stupidly, and he didn't know where to put it.
When she came today, she was mentally prepared to "let the other party account for a quarter of the shares and raise $1.5 billion".
In that case, the present value of station N is equivalent to 4.5 billion, which is about the same as Youku.
In Zhou Tianyin's view, after the dismantling, Station N can be sold as expensive as Youku, and he is already very content.
But Feng Jianxiong just thought about the progress of cutting-edge science and technology, chatted with Jack Ma, and talked about an extra $500 million.
"What he just said is worth about $200,000 per word. Zhou Tianyin silently estimated in his heart, and completely threw himself to the ground.