If you cheat, money will follow

Then, Li Fei demonstrated the performance of the model to Chang Le.

Li Fei wrote a sentence on the paper with a pen:

“How to solve simultaneous equations with quadratic terms?”

Well, it’s just a sentence, there is no specific equation.

He took a photo of the paper and pointed the photo at the model's camera.

Almost instantly, the language model gives the answer.

Answers include, specific example equations, solution procedures and analytical explanations.

Clear and concise.

"Has the model performance reached this level?" Zeng Xi asked in surprise.

"Yes, Mr. Zeng. From the answer to this question, it can be seen that the language model has the ability to recognize fuzzy language." Li Fei said:

"This is very important. Based on this, even if it is unclear or vague questions, it can give corresponding exact answers."

"Of course, in the known world, it can perform even better."

"But it's still very immature in the unknown."

Li Fei wrote another sentence.

Then take pictures and model recognition.

This time, the model gives the theorem content, proof process and proof method involved in the problem.

Similar to encyclopedia, it does not meet the requirements.

Because Li Fei asked the model to use a new method to prove Kummer's theorem.

new method?

This is uncharted territory and beyond the capabilities of the model.

Currently, it does not involve the creative abilities of logical thinking.

"Even so, its performance is already very good." Chang Le said:

"Mr. Ma contacted me yesterday to discuss the commercial use of GPT."

"What do you think?"

"Boss, we have been considering this issue recently. If it is only TOB, then the model cannot meet the higher-level needs of enterprises and institutions." Li Fei said:

"If it is targeted at individuals, it may cause a series of problems such as illegality and crime."

"Rumors, false information, sex, violence, etc. will flood the Internet."

“Language models are very good at counterfeiting.”

"So we recommend opening it up to individuals within a limited framework."

"Can technology realize it?" Chang Le looked at Hinton.

"Technically it can be achieved. Through code, keyword control, and strengthening supervision through weak models, this can be achieved." Hinton said:

"In fact, we have been working hard in this regard and have achieved very good results so far."

"We also believe there is an option to launch the right range of language model products at the right time."

"It's going to be a worldwide sensation."

It will indeed be a sensation.

Because both Google’s general model and OpenAI’s language model have just been iterated to the second generation.

His abilities are very immature and full of loopholes.

Chang Le suddenly looked surprised.

It is rumored that all three hope that the language model can be made non-profit.

Sutskvi explained: "Boss, this is misinformation from the outside world."

“Actually, we are not opposed to making profits. We are opposed to allowing models to run unfettered and unrestricted for the sake of making profits.”

“The blind pursuit of profit maximization will inevitably ignore the safety of the model.”

"We hope to use it commercially under the premise of safety. After all, it is a tool to serve mankind and should not become a spreader of crime."

"That's a good point. I've always been worried about your thoughts and concerns, so I never mentioned this matter to you." Chang Le said with a smile:

"And now that you're supporting it, I'm so happy."

"In this way, we will further optimize and adjust the language model before the year, and we will launch the product after the year."

At this time, Liu Chaoyang answered a phone call and looked happy.

He came to Chang Le and said softly: "Boss, carbon nanotubes are successful."

"Succeeded?" Chang Le was overjoyed, and Zeng Xi was even more overjoyed.

"Yes, it worked."

Materials Building.

Lin Li introduced to Chang Le, Zeng Xi, and Liu Chaoyang and personally demonstrated the carbon nanotube preparation method developed by her and her team.

"We developed a relatively special solution."

"Immerse the carbon substrate into the solution, and the solution will automatically separate layers of carbon nanotubes."

"After the carbon nanotubes on the carbon substrate are completely separated, all that remains are impurities from other components."

"Under a scanning electron microscope, you can clearly see the regular shape and orderly arrangement of carbon nanotubes, and they are all semiconductor in nature."

Lin Li further said:

"This method can produce semiconducting carbon nanotubes in large quantities, neatly and homogeneously."

"What if we need carbon nanotubes with conductive properties?" Liu Chaoyang asked.

"It can be done by adjusting the solution ratio." Lin Li said.

"How high is the purity?" Liu Chaoyang continued to ask.

"Ten nines..." Lin Li said.

"So high?" Liu Chaoyang was surprised.

"It has been repeatedly verified that the purity is between 10 9s and 11 9s. I personally think it can be further improved as long as the concentration of the solution is increased." Lin Li said with great certainty.

Only carbon nanotubes with a purity of more than 6 9s can produce high-performance carbon nanotube chips.

The introduction and demonstration process just now by Lin Li was very simple.

Butenediol, trichloroethane and a specially prepared solution are mixed with each other.

Butenediol, 26 yuan/kg, 2600 yuan/ton;

Trichloroethane, 30 yuan/kg, 3000 yuan/ton;

Obviously, special solutions are not simple.

"How much does this special solution cost?" Liu Chaoyang asked.

"This kind is more expensive. One kilogram costs about 1 million yuan, and this bottle costs 100 million yuan." After Lin Li finished speaking, she listed several configuration materials.

Liu Chaoyang nodded after hearing this: "It turns out they are all rare materials, no wonder they are so expensive."

"But it can be kept for one month and can be reused within one month." Lin Li said.

"Then the cost is not unacceptable." Chang Le nodded.

"Well done, Dr. Lin. I'll give you and your team a credit. The Yangtze River Scholars won't be able to escape. In addition, I will help you actively strive for academicians of the Academy of Engineering." Liu Chaoyang said.

Shu Zhijie's academician application has been submitted. No one will object or dare to object, and there is a high probability that he will not be able to escape.

"You really need to make a big mark, Dr. Lin, because you changed the pattern of an industry." Chang Le said.

"Thank you, boss, Mr. Zeng, and Dean Liu. It finally worked." Lin Li nodded happily.

"Dean Liu and Dr. Lin, the material solution will be handled in accordance with the confidentiality rules of the institute and will be notified by the company," Zeng Xi said.

"it is good!"

The biggest obstacle for carbon-based chips is the preparation of high-purity carbon nanotubes.

Once this problem is solved, the application of carbon-based chips will be able to gain industrialization wings and fly into the homes of millions of ordinary people.

As for the research and development, design, and manufacturing of carbon nanotube chips, including equipment, it can basically continue the silicon-based chip because the carbon nanotube structure is highly similar to the silicon-based chip structure, and the manufacturing methods are also very different.

Eight major processes including glue development, photolithography, etching, ion implantation, oxidation annealing, cleaning, physical vapor deposition and chemical vapor deposition are also required.

No one is missing.

Let’s talk about the difference.

That is, compared with silicon-based chips, carbon nanotube chips have a higher operating frequency, with a limit of 100GHZ.

The operating frequency limit of silicon-based chips is 10GHZ, a 10-fold difference.

The power consumption of carbon nanotube chips is only 1/5 of silicon-based chips.

Therefore, the performance and integration level of carbon nanotube chips using the 90nm process are equivalent to or higher than those of silicon-based chips at the 24nm technology node.

The performance and integration level of carbon nanotube chips using the 28nm process are equivalent to or higher than those of silicon-based chips at the 7nm technology node.

By analogy, if it is a carbon nanotube chip with a 10-nanometer process, it is basically equivalent to a silicon-based chip at the limit of Moore's Law.

3nm?

2nm?

1nm?

According to the research results of the Kyungseong University research team:

When carbon nanotubes reach the theoretical limit (1 nanometer), they can naturally overcome the short channel effect.

The short channel effect, in layman's terms, means that after the Moore's Law of silicon-based chips reaches the top, there will be leakage, voltage instability, and current instability.

Therefore, carbon nanotube chips do not need to use complex three-dimensional transistor technology such as FinFET (Fin Field Effect Transistor).

At the same time, the manufacturing process of carbon-based chips is relatively simple compared to silicon-based chips.

According to a Stanford University research report, if a carbon nanotube chip is designed into a three-dimensional structure, its performance is 1000 times that of a two-dimensional silicon-based chip.