Ted Jiang tells a story comparable to the science fiction novel he wrote.

"I heard this from a computer scientist friend David Krisel, who received a request from a construction company in Titanium, saying that there were some strange problems with their company's Xerox copier. When they were copying, they were copying.

When a house is floor plan, there will always be a slight but unnegligible difference between the copy and the original. In the original, each house has a rectangle to illustrate their area, the rooms are 14.13 square meters and 21.11 respectively.

square meters and 17.42 square meters. However, in the photocopy, all three rooms are marked as 14.13 square meters.”

Liu Yukun had heard this story and showed a mysterious smile.

Li Rui and Dalula felt incredible. Shouldn’t the copy machine just copy the original one? Why is there such a strange difference between the copy and the original one?

If it isn't for the internal program, it's just that the copier has become a tinker?

Tedjiang continued: "The company conducted a comprehensive inspection of the copier and found no problems. In the end, it could only turn to David. David went to see it and found that it was a digital scan document.

A modern copier that produces image files, and between scanning and printing, the copier internal program will compress each digital image file, and the problem occurs in the compression process."

"We all know that compression is divided into two types, one is lossless compression, and the other is lossy compression. The recovered files of lossless compression are the same as the original files, while the recovered files of lossy compression are only approximate values ​​of the original files.

Xerox Copier uses a lossy compression format called jbig2, a very space-saving technical format designed for black and white images. The copying machine recognizes similar areas in the image and takes all of this type.

The area is stored as a copy. When the file is decompressed, the copy reuses the copy to rebuild the image. As a result, the copy determines that the labels for the specified room area are very similar, so it only needs to store the 14.13 square meters of room, and

When printing, reuse this one label for all three rooms.”

"So that's it." Li Rui and Dalula understood.

It seems that it is not because the copier has become a good product, or because there is a problem with the design of the internal program.

It seems that there is no direct connection between a copier's suspected insulting story and artificial intelligence. The compression technology used inside the copier is the key to this story.

Tedjiang continued: "David pointed out this problem and contacted Xerox engineers, who would probably solve the bug with patches soon, and I thought of other problems from this story.

We have always believed that a copy machine is just a complete copy of the original, but in the digital age, it has adopted a subtle and vague way to solve the copy problem, and the copy it makes looks accurate, but is not actually accurate. The human being we are working on

Can intelligence be developed in a similar way?”

"Develop an inaccurate copy?" Dalula asked in surprise.

Ted Jiang smiled: "Of course not."

He paused and explained: "As far as I know, there are other companies in Microsoft and Google that are studying language models, hoping that humans and machines can communicate and talk in some way, which requires the preparation of an extremely large database for machines.

It is best to be able to store the entire World Wide Web for machine calls.”

"That's impossible." Dalula shook her head.

Tedjiang said: "It is indeed impossible, but if you use compression technology, especially lossy compression technology, it is still possible. This requires writing a lossy algorithm to identify the statistical rules in all data, and you can put all the results.

Data information is stored for machine learning and use. The only problem is that because the data is highly compressed, the machine cannot find information by searching for accurate references, and cannot get an exact match, and can only respond through some key points.

”

"chatgpt... no, it should be a question of the truth..." Li Rui murmured in a low voice.

"What did you say?" Ted Jiang asked.

Li Rui shook his head and said, "It's nothing." There were huge waves rolling in his heart. It turned out that someone had proposed the concept of chatgpt like this early.

Tedjiang continued: "I believe that through similar technologies, humans will get a conversational AI in the near future. It retains most of the information on the World Wide Web, but it cannot answer a question accurately, and

Only an approximate value can be found in the database, and then the answer is displayed using unique techniques to create syntax text. This answer will make you feel that it is not bad, but if you think about it carefully, it actually cannot stand scrutiny. This is why artificial intelligence will compress

,The search and display key problems encountered in the display process, whoever can solve these problems first will be able to take the lead in this field."

Ten years later, when the first batch of artificial intelligence models appeared, some of them were shocking, such as chatgpt, and some were embarrassing, such as asking the heart and saying things.

In his previous life, Li Rui didn't know what the gap was. Now that he has heard Ted Jiang's explanation, he has probably gained some understanding.

I don’t know which link has the problem with the messy answers of Wen Xinyiyan, but chatgpt does have unique technologies in data adoption and text creation, so it’s no wonder that everyone was stunned when it came out.

Liu Yukun said, "My idea is that humans should always believe in themselves more than artificial intelligence that is compressed and copied."

Li Rui nodded.

chatgpt is indeed amazing, but there are still a lot of questions in the first few versions, and many questions seem to be answered very well, but if you carefully collect information and compare them, you will find that there are many errors and omissions.

People with normal thinking ability will only regard chatgpt as an auxiliary tool.

But will anyone take its wrong answer as the correct answer?

Just like the history we read, it is actually a text that is reconstructed after lossy compression. It may be right, or wrong, or made up, but it has been regarded as a guide many years later... As for the real

What is the history of this, and who knows what we have lost in this process of compression and reconstruction?

Li Rui benefited a lot from the conversation with Ted Jiang, and he also truly realized the foresight of science fiction writers.

Think about it carefully, is Tedjiang's current view and predictions about artificial intelligence very much like Verne's fantasy and predictions about submarines, traveling around the world, rockets and many other technologies that seem ordinary now?

The literary works written by Verne in that era were fantasies, and those fantasies were realized one by one by scientist in the years to come.
Chapter completed!