【Broadcast】Pay attention to the Biquge www.biquge.info study, get the first-hand news of 515 red envelopes, and the students who have not grabbed the red envelopes after the Chinese New Year can show their skills this time.

"Let's talk about the impact of the online education model on traditional basic education. Or, what do you think is the model you are now promoting on video sites that can continue to be improved and combined with new technologies on the Internet? I don't want to look at the flat material, I just want to listen to the most pertinent questions. ”

Gu Mojie casually glanced at Salman Khan's plan a few times, then threw the thin pages on the coffee table, and tapped his fingers to ask.

He needed to make sure that Khan had dry goods that were worth investing millions of dollars a year in it.

The khan didn't take it seriously, and said directly:

"I don't think I'm the first to do video teaching — when I was teaching my cousin middle school math on video, MIT itself had free online video math lessons. But it turned out that MIT's classes were not as popular as mine, which suggests that there is something wrong with their course mode.

I don't come from an education major, so I know my strength. In terms of teaching ability, I would definitely not be as good as the professors at MIT. In the case of face-to-face interactive teaching, the quality of their teaching will certainly surpass mine. And now I have been overtaken in the field of video lessons, which can only show that they have not developed their strengths and avoided weaknesses in the application of video technology. ”

Gu Mojie gave an encouraging look: "Go on." ”

Khan took a sip of cherry juice and continued: "First of all, the biggest flaw of traditional video teaching is that it focuses entirely on 'teaching', not on 'learning' - that thing is arranged from the perspective of the instructor, not from the perspective of the student. When the teacher said it, he naturally felt that the student understood. Even if the students don't understand it, they can't ask questions, raise questions, or discuss in the face of a recorded video, so they can only copy and accept it.

Of course, there are also cost savings in video teaching compared to ordinary teaching. In addition to allowing a teacher's teaching results to be replicated countless times and heard by more students, it can also ensure repeated listening and selective listening. If the student gets distracted or doesn't understand the content once, the student can pull the progress bar back. Listening to it a few times can deepen the impression and help you grasp it.

The set of video course technology I have explored is different from the simple playback on YouTube, but fully combines the 'deep learning algorithm' jointly discovered by you and Professor Jeff Hinton - my video content push is with a certain intelligence and learning, and sets up a relatively complex structure of multiple video tree and mesh cross push.

For example, if a student does not understand a knowledge point, he or she answers the wrong question at the end of the video. Then I ask the student to click on the symptoms of the error - for example, the correct answer is A and he chooses B, and then I will jump to a small video of 'Why is B wrong, and why did you choose this wrong option because you didn't have a solid grasp of the knowledge?' In the same way, if he chooses C incorrectly, he will also skip to a commentary video.

In order to achieve these goals, my public lecture videos are relatively short. I don't have to set up a video for 45 minutes or even 1 hour as a lecture like MIT or Harvard Business School classes. My class may only be 15 to 20 minutes. But it is enough to explain two or three small knowledge points thoroughly, and leave the exercises at the end. If you get it all right, there's no need to listen to it again. If you do something wrong, then selectively jump to the part where you are wrong.

I think that's what really is teaching students to their aptitude, allowing them to spend their time and energy on things they don't understand yet. Instead of wasting time on a 45-minute lesson from the first minute to the last minute, no matter what he mastered, that would be a waste of a student's life. ”

From a flat and straightforward account video, to a specific analysis of specific problems according to the mastery of knowledge points, the intelligent ascent in this is dù. It's not the half-star that the layman sees.

For Khan's deduction, Gu Mojie immediately felt the huge prospects behind this technical idea.

Perhaps later generations of netizens have seen too many functions such as "guess what you like" and "push of like-minded works" in the era of large numbers and cloud computing applications, and feel that these are not difficult. Even a certain point will be pushed, and even those browsers that specialize in piracy will play the slogan "No matter how much you give, you might as well understand me".

Coupled with these promotions under the slogan of smart push. In fact, it is often polluted by bidding rankings and made into half-hung water-injected pork, so domestic netizens don't think that these push algorithms have anything awesome.

But in fact, if it can not be polluted by the spam information stuffed with money, a mature content screening and push algorithm is a great way to improve user stickiness for a content provider website.

For example, if a station's "large number jù push based on depth algorithm" is really done, book friends should not shout "why is there a shortage of books again", even if it is a niche demand, at least it should be automatically screened out by the system to see and like to read. Rather than sifting through them directly by category and tags, you end up with a bunch of smelly headline parties.

If a music or video website does a good job in this regard, it can theoretically provide a list of books, cloud music playlists, and video pushes that are more suitable for users' tastes......

In the era of the explosion of knowledge, the need to "get knowledge" is no longer scarce, but "finding the number and service you really want without being disturbed by knowledge garbage" has begun to become a luxury.

Efficient retrieval is much more important than hoarding solid knowledge. Only those who are about to be eliminated by rote will be in the next form.

Thinking about this qiē, coupled with the strong core values brought by his own rebirth, Gu Mojie felt a burst of beast blood boiling.

Khan's project is worth the sponsorship.

Gu Mojie's eyes were so keen, and while he was delighted, he asked directly: "However, what you said, it shouldn't be possible to achieve it perfectly with the current technology of the video website, and many settings need to be done manually." The internal logical relationship between many pushes. It is all set manually, and is not based on the depth algorithm and the automatic statistics and induction of large numbers. ”

Khan was slightly embarrassed: "Of course there is still a problem, at present, the exercises left at the end of each teaching video, if the students do it wrong, the specific jump to which follow-up commentary video is the association that I manually set." On the one hand. I certainly don't know enough about depth algorithms to make a set of algorithms that work based on this application.

On the other hand, after all, the maximum number of students who have watched my video classes so far is 100,000, and this sample size is not large enough, and it is really based on the architecture of deep algorithms, perhaps because there are not enough materials for deep learning, and the push effect of the guide is not as good as the current manual setting. ”

Gu Mojie also thinks deeply of this statement.

Algorithms are important to identify the power of a cloud-based network-based AI. But the most important thing is the number of users and the frequency of user use.

This is also why after Google grew into a behemoth, there is no company in the world that can compare with Google in the savage growth of artificial intelligence - no matter how much money you invest, no matter how unscrupulous you are in scientific research, at best you can make an algorithm that is two or three times more efficient than Google's algorithm.

So what? Google's user count and frequency product are five or ten times more than yours, and you have three times the efficiency of the algorithm, and you are still being thrown farther and farther away. Moreover, in the absence of generational technology, it is impossible to have an algorithm with three times the efficiency.

Gu Mojie thought for a while and asked Khan a peripheral question: "You used to do a video open class." Have you received any other charitable donations? ”

Khan thought for a moment and said, "Yes, it turns out that I have been doing it for more than a year and a half. Last year, I received more than $200,000 in donations, and I spent most of it on the programmatic side because I couldn't handle the entire video delivery structure on my own. ”

Gu Mojie has a bottom in his heart.

"What are you afraid of, you only rely on the investment of more than 200,000 US dollars a year, plus your personal unpaid labor, to make the current foundation. Starting this year you can get five million dollars a year. Nothing can be done.

If you don't have enough algorithm engineers, I'll just tune it up for you from the Hatsune Group. Dr. Yan Lei, do you know you? That was the first batch of PhDs in deep learning algorithms brought out by Professor Jeff Hinton, and there was no such major on the earth before. He's been following me at Hatsuingan, and now there's a dozen or so Jeff Hinton and Professor Bengil's disciples with me. All very well done.

If the number of user samples is not enough, you can invest money in public service advertising promotion, so that these courses can get more drainage channel entrances, and millions of dollars a year in advertising costs, I estimate that it is easy to expand the user scale of this business dozens of times.

And I estimate that if all these things are done, it will cost about $3 million a year. There are nearly two million left, and I am counting on you to do more - to spend on course translation and overseas promotion, and I hope to see Khan Academy's courses not only available in English, but also in Chinese, French, German, Russian, Japanese, Italian, Spanish, Portuguese, and other Chinese languages. There is no hurry, you can add three or five language translations every year, and it will take a few years to get it completely solid. If you run too fast, it is estimated that the broadband network is not yet widely available in the backward areas where Spanish, Portuguese and Chinese are spoken. ”

Khan was greatly encouraged by this, but still had a trace of doubt, and said without vomiting:

"Mr. Gu, thank you very much for your generosity. However, I have to remind you of one question - if the course is translated into multiple Chinese, I wonder if this fund can still be used by the federal government to approve the public welfare ** charity fund? Federal law, however, requires that all charitable contributions eligible for tax deductions must not only be located in the United States, employ all U.S. citizens, but must also ensure that "all users are used in the United States."

If it is a project that benefits overseas, it will not be able to enjoy the tax credit for this part of the fund. I don't study law, so I don't know about these things, but last year I thought about the translation course, but I ran into this nail, so I was very impressed. ”

Gu Mojie thought and said, "Is there anything else?" You don't have to worry about legal matters, I'll consult someone else and find a way to solve it when the time comes. You can do the translation thing, I may also give you some ways to reduce the cost, but if the money is not enough, I will take the money directly from the company to you, and I will not try to deduct the tax. ”

ps. Chasing more children's shoes, are there any free appreciation tickets and pen fun pavilion www.biquge.info coins~ The countdown to the 515 red envelope list is over, I'll pull a ticket, ask for an increase and appreciation vote, and finally rush a handful! (To be continued.) )