As a mage, I just want to pursue the truth

Chapter 360: M Language and the First Occupation to be Eliminated

Although Singapore seems to be an independent country, it still depends on America to some extent.

The official website of Americanin states:

The two sides have established a broad and enduring relationship based on common economic interests, defense and security cooperation, and enduring people-to-people relations.

In recent years, America has actually been strengthening its diplomatic relations with Lion City.

Among them, a fact sheet on strengthening the strategic partnership with Lion City was released in 2021, and this point was repeatedly emphasized during the Vice President of America's visit to Lion City that year.

From 2021 to 2031, the relationship between the two sides has been strengthening.

The Li family sold Lion City to Zheng Li, which was unexpected by the American side. By the time they knew about it, the Li family had already completed the letter of intent negotiations with Zheng Li.

The intention has been discussed, and it is too late to stop it now.

Moreover, for American, Lion City is not that important compared with the possible future benefits it may gain from Zheng Li.

Therefore, they remained on the sidelines of the deal.

And if the Li family can make a deal with Zheng Li, why can’t Americain?

Under the manipulation of the family that truly holds power in America, Zheng Li used three potions that could extend life by ten years to gain control of all of America's personnel in Lion City.

At the same time, the previous agreement on leasing a military base in Singapore was abolished, and since then, America's interests and organizations in Singapore have been completely withdrawn.

The Singapore councillors, who had originally thought they could use some American influence behind the scenes, instantly lost their temper.

America has already withdrawn its troops, so why are you still making trouble?

As for contacting the Lion City’s armed forces to launch a coup, this is pure wishful thinking.

Merlin's image in the outside world is not a real entity at all. At the same time, the artificial intelligence that takes over Lion City, how can your armed forces kill the artificial intelligence?

Furthermore, this is the result of a decision made by all the people. You have no legal basis for it, and there is no actual physical goal at all.

Besides, the armed forces of Singapore are not stupid. Would they gamble their future or even their lives for the interests of politicians?
The remaining more than 100,000 civil servants in Singapore were divided up by Zheng Li.

Zheng Li used the name of Lion City to establish several new state-owned enterprises.

These companies will recruit from the general public sector.

If you are obedient and honest, take part in the selection. If you are not obedient and honest, you can make trouble. Zheng Li expressed an attitude that there is no room for compromise.

At the same time, the benefits of the unemployed will be greatly improved, and money will be directly distributed according to the minimum guarantee.

After this combination of measures, all the unemployed civil servants became obedient.

Everyone knows that Merlin would not do anything shady. Since it is a selection exam, these best-of-the-best civil servants naturally feel that they have a chance.

Moreover, for them, the state-owned company established by Merlin has a bright future, whether in terms of treatment or the technology demonstrated by Merlin.

In the interim, Lion City released a series of artificial intelligence frameworks, which read:

“The AI ​​frameworks announced below follow the open source spirit that has been popular since the rise of Bluestar Internet, and will make public some AI frameworks that are better than Bluestar’s existing AI frameworks in some aspects.

These AI frameworks are free for all individuals and businesses to use.”

In Zheng Li's opinion, Bluestar's productivity is not improving fast enough, and its progress in integrating artificial intelligence with other industries is still not as good as expected.

After flexing his AI muscles in Lion City’s government-oriented AI, he hopes to further advance Blue Star’s AI and industry integration progress by making the AI ​​framework public.

He can complete the entire process himself, or he can bring some inspiration to the field of artificial intelligence on Blue Star and let them complete it under his guidance.

With the further implementation of artificial intelligence and combined with biophotovoltaic technology, Bluestar’s employment population will be reduced by 80 percent or even more.

A large amount of productivity has been released, whether it is to engage in more technically demanding work or to enrich the virtual world ecosystem, it is better than doing some meaningless work now.

Zheng Li has been in Blue Star for quite some time. Although he has been at the top of society for more than ten years, he has gained a certain understanding of the social ecology of Blue Star through the Internet.

In his opinion, many jobs on Blue Star are meaningless and exist purely to torture people without any output.

At least the manufacturing and service industries have output.

Even if it is a productive job, Zheng Li still feels that it is meaningless. Intelligent life that has been born with great difficulty in the vast universe should not spend its life on such boring mechanical labor.

Intelligent life needs to be allowed to find the meaning of its own existence.

Instead of overcoming the many hurdles designed by society, even if you stumble through all the hurdles, you will still feel that your life has been wasted.

For Zheng Li, Lion City is just a test field, and it will eventually be promoted to the entire Blue Planet.

The AI ​​framework that Zheng Li publicly released under the domain name magic.com includes a new programming language and open source packages around this programming language, that is, tools around this programming language.

The reason why Python has become so popular in recent years is closely related to the fact that it has a large number of useful toolkits.

These packaged packages greatly reduce the difficulty of programming in Python.

As soon as the M language released by Merlin was announced, it quickly caused an earthquake in the entire Internet industry.

"You are too modest. It is not just that he is excellent in some aspects, but that he is excellent in all aspects.

Although I haven’t figured it out yet, I always feel that this thing is much more advanced than the mainstream artificial intelligence framework on Blue Star.”

The main reason is that this thing is too professional, and things that are too professional are destined to be isolated from the mainstream group.

Even professional algorithm engineers need some time to study.

Most of the netizens' discussions were concerned about whether they would lose their jobs:
“It feels like Merlin is not a magician, but a programmer, trying every possible way to promote the large-scale implementation of artificial intelligence.

First, artificial intelligence was used to replace manpower in Lion City, and then open source AI programming languages ​​and frameworks were released.

I seriously wonder if the way magicians use magic is a bit like programming?"

"Merlin, please stop using your magic. I finally found a job, but I hope you won't demote me again. I don't want to graduate from Weibo.

This is the dream job I finally found.”

The person who posted the following Weibo post is a pornographic film appraiser.

Such a career does exist, and the pay is pretty good.

But for him, it was a prophecy that came true.

After Merlin released the AI ​​language and supporting frameworks and toolkits, these Internet giants asked their algorithm engineers to put down all their work:
“The only job for everyone during this period is to learn how to use the M language, M framework, and the accompanying packages.

Then we combined our work to see where we could use M language to rewrite it. "The first thing that was used was AI pornography detection.

AI porn detection has always been a core demand for content security. Basically, everyone from QQ, Tencent, Weibo, and ByteDance are researching this thing.

It is a classic that is easy to get started but difficult to master.

In the early days, pornographic film identification was basically done through manual review, which was a labor-intensive job.

Later, as the number of Internet users increased and the content increased, the cost of manual review became increasingly high, so the use of AI + manual methods to identify pornographic content became the mainstream.

The AI+manual approach generally involves using a machine to filter out most of the images that are definitely normal and those that are definitely problematic, and then handing the rest over to manual review. This can significantly reduce labor costs, and the better the machine recognition effect, the lower the manual review cost.

AI porn detection is actually a relatively broad concept, which can be implemented through a rule system, such as setting up a blacklist library based on MD5, user IP and other information, and directly intercepting based on rules.

Most of them still use algorithmic models, that is, use algorithmic models to determine whether an image contains sq information, which is essentially image recognition.

Image recognition currently even surpasses humans in some tasks.

The most common algorithm in image recognition is image classification algorithm, from AlexNet to VGG, from ResNet to DenseNet.

The current image classification algorithms can distinguish ImageNet's 1000 categories of data relatively accurately. Pornographic identification itself is also about classifying input images, so it is natural to use image classification algorithms.

Moreover, the target detection algorithm can be used to detect the dew point in the SQ image, which is also a relatively reliable means.

In addition, there are features and logic constructed at the business level, such as whether there is a person, the area of ​​the skin, etc., which are used to assist in judgment, and are indeed effective in some cases.

The difficulty of AI porn detection lies mainly in soft videos that do not reveal any nude parts, videos with small features, non-universal videos, and cartoon and anime videos, etc.

This is the difficulty of using AI to identify pornographic images, and the difficulty of using AI to identify pornographic images through videos and audio is even greater.

Moreover, for these Internet giants, even if they can achieve a 99% interception rate, the remaining 1% of the content will be of great concern.

Taking Weibo as an example, the data generated every day is measured in T.

Even if there are dozens of Ts, a one percent interception error rate is enough to make the team very upset.

What’s more important is that China’s content review industry is not just limited to pornographic content detection. OCR review has been done by major Internet companies for a long time.

Click till.

By the same token, a 99 percent interception rate is unacceptable to them.

Therefore, until 2031, the method of pornographic identification will still be a combination of AI and manual labor.

It just means that in 2021, a company the size of Weibo may need thousands of pornographic inspectors, but by 2031, it will only need hundreds.

A four-digit number becomes a three-digit number.

As for WeChat, every picture or video you send on WeChat will be OCR'd and then sent to the back-end review interface.

A weight algorithm is used to determine whether you have violated any rules. Once a specific rule is triggered, your weight will be determined. Once the weight exceeds the threshold, you will be the focus of attention.

A human will review your application.

Of course, this kind of review is not limited to large domestic companies. Facebook, Instagram, YouTube, Google and Twitter all have large manual review teams.

Their audit team is based in the Philippines.

In 2018, PBS aired a documentary about this incident.

For domestic and foreign Internet giants, they can achieve close to 98% accuracy in text recognition, but in image classification tests, they can only achieve around %.

Moreover, it has very high requirements on computing power and cannot be used in actual production environments at all.

This is the result of ImageNet's annual image classification test competition. Image and video recognition in actual operations is much more difficult than the ImageNet competition.

After Zheng Li made the AI ​​algorithm public, he wrote the model using M language.

From deployment to use, this has broken through the cognition of these Internet companies. An AI model that can achieve a 99.9% recognition success rate for content requires about the same computing power as before.

In other words, the scale of AI porn detection, which previously employed hundreds of people at major Internet companies, has now been reduced by another 90 percent.

AI porn detection is just the first application of M language.

The AI ​​technology it represents has greatly improved its computing power advantages and the difficulty of implementing the algorithm itself.

It allows most of the originally complex architectures to be easily constructed and can be widely used in the industrial field.

A large number of easily composable modular components are perfectly adapted for writing your own layer types, computational graph abstraction, data and model parallelism, dynamic type stability, etc.

After being used by algorithm engineers, M language-related projects on Github became popular in the community.

In foreign programmer communities, active experts are complaining, and some experts who rarely speak up have also popped up:
“Merlin’s technology is far beyond our imagination. If he developed the M language himself, then his achievements in the field of AI are greater than all of us.

It is really rare to find an AI programming language that is free of any flaws.

Caffe, for example, can port Matlab's fast convolutional network implementation to C and C++, which is suitable for feedforward networks and image processing, and the model can be trained without any code.

However, Caffe cannot be used for recurrent networks and has very poor scalability.

The TensorFlow framework launched by Google can generate computational graphs and then perform automatic differentiation, eliminating the need for manual coding when trying new neural network arrangements.

But it runs very slowly and is very error-prone in large software projects.

Basically, the AI ​​frameworks launched by major Internet giants in recent years have more or less problems.

Of course this is unavoidable, nothing is perfect.

Just like no matter how many people use Java or how long it has been active, shortcomings can still be found.

But in the field of artificial intelligence, M language seems to have no shortcomings.

It seems to have reached perfection after countless tests and optimizations.

This makes me seriously doubt that the artificial intelligence that Merlin possesses is strong artificial intelligence.

He can realize almost all of the Blue Star people's fantasies about artificial intelligence.

Including self-programming.

It doesn't seem so strange that one piece of code can compile another piece of code.

After all, today’s AI models can already perform simple programming on their own.”

(End of this chapter)