Unprecedented popularity means great advertising opportunities.
Meng Fanqi is currently intensively testing the performance of the new face algorithm and the recognition success rate of generative fake faces in his company.
Originally, this matter only needed to be done roughly, but now the security issue of generated content is highly concerned, and there is no room for any mistakes in this regard.
While Meng Fanqi is striving to improve the effect of his algorithm, facial recognition teams around the world are thinking hard.
Opportunities and challenges coexist. Although the entire face recognition direction faces a huge difficulty in distinguishing authenticity, as long as this problem is solved, the benefits will be equally huge.
This kind of security risk is a fundamental problem that must be solved. It can be said that whoever can solve this problem will master the core technology and be able to pick the biggest fruits first.
Facebook's DeepFace team had just iterated a round based on several new data from Meng Fanqi. Before they could recover from the joy of achieving several key improvements, they directly encountered a major crisis.
All the brothers were numb.
One second ago, he felt that Meng Fanqi was the savior of his team, but now he felt the other way around, feeling that Meng Fanqi was the team's ultimate enemy.
"Why did this birdman make technological breakthroughs so quickly..." Yang Ming sighed deeply. He had a team of four who focused on human faces. They were very familiar with the experiment, and the division of labor and cooperation were even more intimate.
Why is it that every time I get close to publishing, I am always interrupted by this birdman.
The last time was a revolutionary breakthrough in the backbone network, which forced several people to iterate a new round of experiments to ensure the status of their technology.
It's better now. Instead of making a breakthrough, we just changed the entire track and solved a revolutionary safety problem for you.
This is bad luck, and it directly becomes a question of whether this technology can still be used.
The entire track was destroyed.
“How accurately does our algorithm recognize generative fake faces?” Wolf asked.
"It depends on the situation. It mainly depends on how much of the fake face you have for it to see."
"What do you think?" Wolf was stunned for a moment.
"Because it will think that all the faces are real anyway, so if you don't show it any fake faces, the accuracy will be 100%. If you show it all fake faces, then it won't be able to recognize anyone." .”
"You're still poor, right?" Wolf's face darkened. How long had he been showing off his wits?
"Sigh...if the security issue cannot be solved, the company's short-term commercial face vision algorithm plan will be postponed to long-term."
Mrs. Wo was also numb. The short-term commercial plan became long-term, which directly meant that the value and status of the four of them were shattered.
The researchers originally did not expect to reach the peak of their lives, but they also wanted to improve their lives, but now they are far away from hope.
Although companies like Facebook and Google are not stingy in investing in technology, whether the technology you create can be monetized, remuneration and status are still far behind.
You are like Meng Fanqi, who was originally an ordinary wage earner at Google Brain. With three swipes of the ax, Google's cash flow was inconvenient to give him that much share. It has already negotiated that a considerable part of it be given in the form of equity.
He instantly became a shareholder with a large share, and his status naturally increased.
This thing is quite outrageous. He has not even officially joined the company yet, and he has accidentally transformed from a wage earner into a small boss.
Although after calculation, Meng Fanqi will be able to obtain less than 0.3% of the Google equity in one year, but Google is a giant that has gone through multiple rounds of financing.
Even founders Page and Brin only hold about 5.5% of the shares.
Google's stock price has been taking off all the way, from around 30 in 2014 to the highest peak of nearly 150 in 2022.
If you need to preserve the value of assets worth billions of dollars, it is also a good choice to hold them first as equity.
When there is an investment opportunity, and then take out a part of it, such as Tesla or Bitcoin, the profit will be dozens of times.
However, these opportunities may not be able to withstand too much capital. Musk will not sit back and watch his crazy bargain hunting. BTC has too many bargain hunting on its own, and capital may also choose other currencies to harvest people.
In short, the purpose is to preserve value in Google first and then diversify the investment. BTC, Tesla, invest in start-ups, start your own business, and bloom more.
This is the safest way.
At the same time, at the Chinese University of Hong Kong in Huaguo Port City, Huaguo scholar Tang Xiaoou and his group of students and researchers looked solemn and the atmosphere was gloomy.
Just a week ago, the face recognition ability of Tang Xiaoou's team reached the level of the human eye for the first time (97.53%). According to the original plan, the team would spend two or three months to push this result forward.
Then the era of facial recognition technology will begin.
Unfortunately, Meng Fanqi intervened and doubled the difficulty of landing.
"Do we currently have any methods to distinguish the content of generative expressions?" Tang Xiaoou is the mentor of Kai Ming, the original author of residual networks. Kai Ming's blockbuster dark channel defogging algorithm in 2009 was developed under the guidance of Professor Tang.
In one fell swoop, it won the best paper at the top visual conference CVPR, which is the first Asian paper to win this honor in the 25-year history of this top conference.
Before deep learning became so popular, Professor Tang's laboratory had frequently published relevant papers at top international conferences.
At the top visual conferences between 2011 and 2013, the number of papers produced by Professor Tang's laboratory in the field of deep learning was equivalent to that of all other institutions in the world combined.
It is enough to show its vision, technical strength and theoretical foundation.
But Professor Tang was still at a loss when faced with this sudden security problem.
"Mr. Tang, although FaceGAN has published papers and demonstration websites, we are completely unclear about its specific training code and details. In this case, it is difficult for us to summarize any valuable rules."
"Yes, Mr. Tang, generative technology is a relatively new technology, and it is the first time it has been used on face images. We have no experience in this area, and it may take a long time to solve this problem."
"If we collect a large amount of generated content through their [these people do not exist] website, is there any chance to summarize any rules?" Tang Xiaoou frowned. The other party's technology was novel and he was completely unfamiliar with it.
It takes too long to figure it out from the source.
His idea is to use these false contents to create a targeted detector.
Separate from your original core identification part and proceed independently.
It has to be said that this is also an idea. The model that can generate these false data is also trained from this kind of confrontation. It is difficult to achieve any results in the short term.
"We will release the technology first and solve this problem later." Almost at the same time, both Facebook and Tang Xiaoou Laboratory made this decision.
In the original timeline, the technology releases on both sides would be delayed by another month or two.
However, the sudden problem made both parties realize that if they don't release it, the situation will become even more embarrassing.
