It was Wang Kai's own idea to start a business with Meng Fanqi, but Meng Fanqi still felt a little embarrassed. After all, he was specially sent to him by Robin Li to learn this part of the code, and he was a little ashamed to abduct him directly.
So the next day came Baidu, which optimized Baidu's current method of identifying license plate numbers from pictures to make up for it.
It is shameful to say that during this period of time in China, he was working behind closed doors, eager to break through technology and accumulate his technical prestige in various directions. He did not consider starting a business, nor did he make any preparations related to starting a business in advance.
If you want to start a business at the last minute, it seems a bit hasty. You will probably spend a lot of time on recruitment interviews early next year.
The first is that there are no technical personnel available around us, and we did not look for them in advance. The familiar classmates are all still studying for undergraduates, so they are not of great use.
Although this is not a big problem in the early days, Meng Fanqi can produce results with all his strength, but to build a company, senior programmers are still indispensable.
After all, Meng Fanqi only knew a little bit about most of the front-end and back-end apps, which was at most an entry-level level.
I know a few of my colleagues from my previous life, not to mention that most of them are still studying. It would be outrageous to approach them rashly like this. I know the other person, but the other person doesn't know me at all.
However, there are a few seniors from the past life who have some level and structure, but are just workers who have just entered the workplace at this time, and can be recruited.
At the moment, only Wang Kai is really available. He is from a major and has experience in a large factory. He has a group of classmates with the same resume who can be brought over. He has interacted with him many times and is considered familiar with him.
"You have to think carefully about starting a business with me at this stage. I can give you more salary, but I can't guarantee whether it will work in the end." Meng Fanqi's ugly words came before: "And after I go to the United States in a few months, In addition to being responsible for the technical aspects, the rest is up to you."
"How can I let you worry so much anymore!" Wang Kai said on the spot that nothing else would be a big deal. Why do I have no motivation when I work part-time? In a word, the company is not my own, so I just take the dead salary and do as much work as I am paid.
The money you give is very little, and you keep nagging me all day long. It’s hard for me to help you with this little money.
But now there is hope to get shares. It is different. The early stage company has taken shape. If it only does the pure software part, it can support Meng Fanqi's AI algorithm to be made into a product, and it still does not require ten people.
As long as the product is made and one or two shares are allocated, this is the starting point of financial freedom.
The small salary of a few hundred thousand from my job is in my early thirties after tax. After excluding daily expenses, I can save up to two hundred and fifty thousand a year.
If you work from twenty-five to sixty-five, you will have saved ten million.
When a start-up company is struggling, the biggest fear is to die before it starts, and all the options will turn into waste paper.
In Wang Kai's view, this biggest risk is not a problem at all and is not worthy of doubt.
With Meng Fanqi's current technical level, he can definitely go all the way to a listing.
Just relying on the advanced facial recognition algorithm itself, it is not difficult for the company to directly sell tens of millions of products.
If there are several complete projects, the valuation can easily go up to one digit.
If we really get involved in government projects and equip customs, government agencies, and transportation stations with these technologies, it will be an astronomical figure that Wang Kai dare not even imagine.
The facial recognition project was something Wang Kai thought about deeply, but in Meng Fanqi's eyes, it was insignificant and something he had almost forgotten.
“There are so many things that can be done. After Google’s recommendation and advertising algorithm updates, I will have to carefully plan and record them.”
At this moment, Meng Fanqi is updating his Google Scholar profile information. It has not been long since he published a large number of papers in Sydney, but it has been some time since he published papers on generative adversarial algorithms.
He wanted to see if his butterfly had caused any big changes.
The most convenient way is to see which papers have cited you and see if there are any significant research results.
After updating his Google Scholar profile, Meng Fanqi couldn't help but be surprised. In just a few days, he already had more than 20 paper citations.
After careful analysis, in fact, these more than 20 citations came from only four or five papers.
Because Meng Fanqi's announcement this time has revolutionized the entire paradigm so thoroughly, and the code is also open source, any current research on deep learning may start by citing many of his articles.
Residuals, optimizers, training methods, and data enhancement, almost no one can avoid these four kings.
For every additional article in the deep learning community, Meng Fanqi's citations will almost quadruple, and this multiple will continue to expand in the future.
As of 2023, the most cited scholars in human history will have a total of millions of citations.
The number of articles in the AI field has rapidly increased from more than 20,000 articles per year in 2012 to approximately 135,000 articles per year in 2021.
If this trend continues, within four or five years, Meng Fanqi will become the person with the most cited papers in history at the age of 25-26.
And in the following years, it continued to grow exponentially.
"Before I was reborn, Kaiming, the original author of the residual network, had been cited more than 400,000 times." Meng Fanqi briefly recalled that the well-known AI technologies he has published and plans to publish are several times that of Kaiming.
It is not impossible to wait until 2023 to exceed 3 million.
Academic papers are written as fast as online novels. The number of citations is equivalent to a subscription. The actual number of times the article is read is dozens of times that.
To be able to write an academic paper with such enthusiasm is unprecedented and unprecedented.
Among these published articles, the Google mailbox left by Meng Fanqi has already been flooded with emails from all sources.
When Meng Fanqi clicked in, the computer froze and it took more than half a minute to recover.
In the inbox, there are people asking for questions, magazines asking for manuscripts, people asking for unpublished code, and colleagues saying hello, everything is there.
Most of these emails were written in English, but some were written in Chinese, which should indicate his nationality.
After scanning around, one of the emails from Shanghai City Public Health Center caught his attention.
Meng Fanqi searched and found that the center is a municipal-level tertiary hospital that has been established for a century. It has many types of large-scale advanced medical equipment and is especially good at the diagnosis and treatment of liver diseases.
Meng Fanqi checked the details of the hospital and concluded that they should have imaging results of different instruments for many diseases on hand.
After reading the email, Meng Fanqi understood the other party's intention. The U-Net segmentation method with residuals that he had updated previously had greatly improved the technical level and segmentation effect of image segmentation.
Especially in the relatively fine-grained object segmentation, there has been a very significant leap.
Segmentation type tasks are very important in medical imaging applications.
Because medical images are usually taken after preliminary diagnosis by various departments.
Classification and detection are not of great significance, and further content analysis is the main requirement.
For example, by carefully isolating the lesion area and assisting in diagnosing the extent of the lesion, it will be easier for medical staff to diagnose and save a lot of time.
"Shanghai's tertiary hospital moved so quickly. It has already noticed me." Meng Fanqi knew the U network's significant impact on medical care, but he thought this would be a task after the year.
But I didn't expect that someone would come to the door the day before yesterday.
Meng Fanqi corrected his attitude and felt that a storm was coming. The two behemoths of the government and the hospital seemed to be close at hand.
It seems that I need to be more serious about starting a business this time.
Not to mention the demand for human faces, a large-scale government project, medical AI alone is enough to be launched on the market.
Meng Fanqi thought about it carefully and felt that it would be best to divide these two directions into two companies.
