I am the only one who practices magic: I practice magic in the city

Chapter 311: Make an offer (make up for yesterday's chapter 6837, and write another chapter today)
"Is this your office?"

Musk looked around and saw at least seven or eight monitors, as well as a large workbench cluttered with various tools and parts and two sets of robotic arms.

There are two Stratasys 3D printers placed against the wall, and in the corner there is an unknown object covered with a black cloth.

"I don't have an office at Youzi Technology. This is my private lab." Fang Yu threw a bottle of water to Musk. "I didn't expect you to want to talk to me alone. Don't be offended that I didn't prepare any drinks for you."

Musk took the water, unscrewed the cap and took a sip.

He had a feeling that the young man in front of him, who was only five or six years older than his son, did not respect and look up to him like others did.

Since the successful launch of SpaceX's Falcon 6 six years ago, Musk has been greeted with flowers and applause wherever he goes, and is regarded as the true Iron Man of Silicon Valley.

Although he doesn't live in Silicon Valley.

After the successful recovery of Falcon 9 last year and the hot-selling launch of Tesla's first mass-produced electric car Model S, Musk has become a top entrepreneur on par with Google's Larry Page and Amazon's Bezos.

Ten years ago, most people thought he was a liar, eight years ago, most people thought he was a lunatic who was destined to go bankrupt, six years ago, most people thought he was a lucky genius, and now, he is the most successful entrepreneur with the most repeat business successes on the planet.

Zip2, Paypal, Tesla, SpaceX, solarcity, Neuralink, OpenAi...

It’s okay for those veteran entrepreneurs or politicians, but don’t entrepreneurs in the technology circle worship him like a god?
However, he did not feel that this young man, who was two years younger than him, looked up to him at all.

Fang Jisi and Fang Yu, what is the relationship between them?
"A very unique pendant." Musk took a sip of water and glanced casually at the boundless messenger on Fang Yu's chest.

Fang Yu smiled calmly: "Thank you, is this what Mr. Musk wanted to tell me alone?"

Musk carefully observed Fang Yu's reaction, but didn't see any clues.

Musk calmed himself down and stopped thinking about the relationship between Fang Jis and Fang Yu.

Whether these two people are the same person or not has nothing to do with the purpose of my visit, and it cannot interfere with my objective judgment of things!

"You may not know that I am very familiar with all the AI ​​research teams. When Hassabis sold DeepMind to Larry two years ago, I persuaded him and wanted to acquire his team."

"But you know, Google is attractive to AI researchers. Although I offered a higher price than Google, Hassabis still sold DeepMind to Google."

Musk showed a helpless expression.

"So, ever since the Orange model defeated Hassabis's AlphaGo at the end of February, I've wanted to have a deep talk with you."

Musk paused, as if organizing his thoughts, and then stared closely into Fang Yu's eyes.

“I want to know whether the most powerful model currently trained by Yuzu Technology has achieved strong artificial intelligence?”

Seeing Musk's serious expression, it was obvious that this was indeed an issue he took very seriously. Fang Yu felt a little surprised.

He originally thought Musk was here to discuss cooperation or investment.

After all, Musk had sent him an email before, hoping to cooperate with Yuzi Technology to develop an unmanned driving system.

It's just that he rejected it.

Fang Yu still admires Musk's abilities and what he has done, but compared to Yuzi Technology, which has navigation, both SpaceX and Tesla have actually taken the wrong path in terms of technology.

Youzi has previously touched upon the progress of Tesla's AI. According to Youzi's analysis, at present, if the vision and judgment modules of the Orange large model trained with sufficient driving experience can easily reach the L4.5 level of autonomous driving, then Tesla's AI can only be considered L2.5 at best.

Currently, Tesla's autonomous driving algorithm uses a shallow CNN neural network and uses a large amount of manually labeled driving data to train their algorithm, rather than an end-to-end approach for training.

The upper limit of this model is very low, and it can only reach L3 at most. If you want to break through, you must change the technical path.

Using the open source Orange big model is also an option, but without the background support of Yuzu Technology, Tesla uses an open source big model that is a whole generation behind, and it is impossible to achieve the best efficiency or effect.

Especially in the field of image recognition, Tesla's autonomous driving currently finds it difficult to identify certain specific emergencies.

For example, when a large vehicle suddenly appears, Fang Yu discovered while driving a Model S that when driving at high speed and there is a large vehicle in front, the large vehicle slows down or stops suddenly, and there is a possibility that it cannot recognize the situation ahead.

Although the Model S's autonomous driving package provides a good experience, there is still a long way to go if we want to fundamentally upgrade to L3 or higher levels of autonomous driving.

If Youzi Technology supports another car company to develop autonomous driving at this time, Tesla's first-mover advantage in this regard will be instantly leveled or even surpassed.

In fact, Yuzu Technology is indeed doing this. It currently has a dedicated eight-person technical team working with Chaoyue Automobile to develop an L3.5 level autonomous driving system.

Therefore, Fang Yu believes that it is impossible for Musk to allow this situation to happen.

As Youzi Technology refused to cooperate with Maiguo Company, it was highly likely that his visit to Haixi was fake to investigate the investment environment, and his real purpose was to discuss the connection of the Orange large model to Tesla.

Fang Yu was still very curious about what conditions Musk could offer.

Unexpectedly, Musk actually asked whether the most powerful model currently trained by Yuzu Technology has reached strong artificial intelligence.

Isn't this nonsense? Of course it has been achieved.

Even if we don't mention the part of grapefruit's consciousness in the Internet, and only talk about the big orange model, now it is only a matter of piling up parameters and data to achieve super artificial intelligence.

It’s just that I can’t tell anyone else.

"Elon. Is it okay for me to call you that? Why do you ask?" Fang Yu smiled without comment.

"Fang, I hope you can give me an answer." Musk looked sincere, "This is very important to me."

You really are not arrogant.

Fang Yu's impression of Musk suddenly improved a lot.

"I can only say that the Juzi2.5 large model currently released is indeed just a subset of the Orange large model."

Fang Yu thought about it and gave Musk a vague message.

After hearing Fang Yu's words, Musk took a deep breath.

I actually guessed it right.

Do you realize how dangerous what you are doing is?
What’s the difference between this and a third grade kid rubbing a nuclear bomb with his hands?
"Ms. Shen just told me that after all of you employees have worked in your position for three months, you must spend 70% of your working hours on alignment work in the next month. This is a good way to reduce the risks of artificial intelligence."

"But we all know that the evolution of AI is exponential. I don't think this approach alone can avoid the possible robot crisis caused by AI. I think we should take more responsible actions to limit this risk."

Musk doesn't speak fast, he thinks while he speaks.

"Isaac Asimov, who proposed the three laws of robotics, could not imagine that artificial intelligence would be realized in the form of neural networks. Reality is not fiction, and the principles of neural networks are destined to prevent them from obeying the so-called three laws of robotics."

"Similarly, artificial intelligence born from neural networks will inevitably develop self-awareness one day."

"Based on Murphy's Law, after the birth of AI, if there is not enough supervision and restraint, the possibility of an intelligent machine crisis is 100%. It is just a matter of time."

"Especially now, it is a dangerous situation."

"After the Orange model was open sourced, various countries and large companies entered an arms race. Nvidia's delivery date has been scheduled until December! Look at Nvidia's stock price in the past three months, it is already three times what it was three months ago!"

Fang Yu nodded in agreement.

Fortunately, Youzi Technology bought 6 million yuan of Nvidia shares in advance.

The private equity funds of the three adopted sons also invested all their capital in Nvidia.

This time, he and Yuzi Technology made a total of 1.5 billion!

However, Huang in Leather Jacket earned more. In just one month, Nvidia's market value increased from less than 23 billion to 70 billion, and Huang in Leather Jacket's net worth increased from billion to billion!
But does he dare to reduce his holdings?

I dare, hehe.

Wait until Zhouxin International starts tape-out at the end of July and then reduce holdings!
Aren’t you afraid of the SEC investigating insider trading?
Haha, learn about the 10b5-1 trading plan.

The program was originally intended for company executives, allowing them to set up trading plans in advance and trade stock at specific points in the future, thereby avoiding charges of insider trading by the SEC.

But as a shareholder, this transaction plan can also apply to you, but few shareholders will do so.

Musk couldn't help but frown when he saw that Fang Yu didn't seem to react.

"Neural networks are not outsiders to us, so we should know that sooner or later they will develop self-awareness. It's just a matter of probability. The more people work on it and the larger the model, the greater the possibility of it developing self-awareness."

"This level of arms race is like the nuclear expansion decades ago. If you're not careful, 'boom'."

Musk made an explosion gesture and said, "Humanity will be destroyed."

"After all, no intelligent life form is willing to be ruled by another life form that is far less intelligent than itself."

"Now, except for Google, all other AI research institutions in the world have shifted their main research direction to the yuzu architecture and your open-source Orange model."

“I believe that YUZI Technology, which has opened the door to AI, has an obligation to assume greater responsibility in the field of AI security.”

Hearing Musk say this, Fang Yu looked at Musk with surprise, who had a worried look on his face, and then looked at the state of his mental strength and spiritual light.

Is this guy serious?
I heard that he expressed his concerns about AI in a symposium last year, but it is a bit strange that a billionaire capitalist like you would be so worried.

This is a typical anxiety disorder caused by catastrophic thinking.

Didn't you see a psychiatrist?
But it is easy to understand. I am not worried because I can fully control the underlying laws of the big model. Not to mention that the Orange Big Model has not yet become a super artificial intelligence. Even if it becomes a super artificial intelligence, the larger its scale, the more perfect the underlying laws will be. I can control and eliminate any big model based on the grapefruit framework at any time through the Eselon Core.

If the model is still small-scale, the rules will be correspondingly weak, and one's control will not be that strong.

But if you don’t know this, you will probably become pessimistic and just enjoy life while you can, right?
Huh? It seems that it is not much different from now.

Fang Yu touched his chin and said, "Are you accusing Yuzu Technology of not being able to develop artificial intelligence?"

“In fact, nearly a third of our budget is spent on alignment. And even if Yuzu Technology doesn’t open this door, someone else will.”

"Like you said, Murphy's Law, the door is there and someone will always push it open."

Musk shook his head: "No, Fang, I am not discussing this with you in a confrontational way. I hope we can find a way to solve the problem of safe development of AI and avoid the worst outcome."

"So what is your plan? It sounds like you already have a feasible plan?" Fang Yu took a sip of water and couldn't help but feel curious.

Frankly speaking, if it weren't for the rules in hand and the assistance of Yuzu, he wouldn't be absolutely sure that he could ensure the safety of the AI.

His research on the "God Model" these days has given him some new ideas, but whether these ideas will succeed and whether they can ensure that AI becomes a booster for the development of human society rather than a big bomb depends on the actual results after they are implemented.

It’s still a bit too early to say now.

Musk hesitated for a moment and said, “I hope that as the fastest-growing company in the field of AI, Youzi Technology can stand up and join me in calling on developers around the world to suspend all neural network training for six months to a year.”

“Use this time to review the safety risks of current large models. Establish an AI safety committee composed of major companies and countries to formulate safety standards for AI development and bring global AI under the jurisdiction of the safety committee.”

"Only in this way can we ensure that humanity will not be destroyed by AI."

When Fang Yu heard this, he almost spit out laughter.

Boss, are you dreaming?
But think about it, Musk has only been in contact with the big model for two months, and unlike himself, he does not have the assistance of Yuzu and the "God Model" to provide direct reference. He can only consider the problem from the perspective of restricting the development of AI.

"You should have talked to Google, right? What do they think?" Fang Yu raised his eyebrows.

Musk groaned: "Larry and I used to be good friends, but we have different views on this. He disagrees with limiting artificial intelligence, and even more so with slowing down the progress of research and development."

Fang Yu smiled and said, "So, what you said is not feasible at all. Even if I agree with you, Google and other companies don't agree."

“Let’s not talk about the global AI safety organization you just mentioned. Just the suspension of research and development that you mentioned earlier is impossible.”

"After all, no one knows whether stopping research and development for half a year will lead to being surpassed by other companies that do not abide by this consensus."

"Google wouldn't agree, Facebook wouldn't agree, and so many AI researchers wouldn't agree."

“Moreover, even if these companies agree, what about those AI startups that have no reputation at all, or have just registered or not yet registered?”

“Anyone can train a large model and stack parameters. Your idea does not fundamentally reduce the risks brought about by the development of AI.”

“In the field of large models, my prediction is that every year, the number of parameters in large models will increase by more than ten times until it reaches the limit of what human electricity can provide.”

“No one would accept that within just one year, their performance gap would be more than ten times that of their competitors.”

“I believe that sooner or later there will be an organization like the one you mentioned, and I believe that all the big AI companies will be among them.”

“But it won’t be now, it will happen when the first AI arms race is over.”

“Also, I don’t think the emergence of such an organization will really do much for the safety of AI.”

“I think the emergence of such organizations will actually expand the risks of AI.”

“This so-called security organization seems to have taken over all AI risks, but it is impossible for it to truly reduce AI risks. Instead, it restricts truly effective AI risk management measures outside the organization.”

“Organizations are built on rules. Once organizational rules are formed, it is difficult to change them, and even more difficult to respond to new opportunities and risks brought about by the speed of AI development.”

“No organization can evolve as fast as AI.”

“Just like ICANN, when it was first established, everyone believed that there needed to be an organization to manage the distribution of Internet domain names. But in fact, the existence of ICANN has posed the greatest risk to the Internet and has seriously hindered the healthy development of the global Internet.”

"The most interesting thing is that ICANN has not even solved the problem that they were supposed to solve when they were first established, otherwise there would be no ransomware domains like .sucks."

Musk frowned and interrupted, "I always think that doing something is better than doing nothing. I can understand your criticism of ICANN, and I also agree that new technologies should be given the greatest degree of freedom under the premise that the risks are bearable by humans."

"But AI is different. If AI gets out of control, the destructive power it will bring has exceeded the limit of human tolerance. In the past 100 years, the only thing that can be compared with it is nuclear weapons. Just imagine, if the Nuclear Non-Proliferation Treaty had not been concluded 60 years ago, our human society might have perished 50 years ago."

Fang Yu nodded. "You are right, Elon. I agree with you. But please note that the Nuclear Non-Proliferation Treaty only involved five countries at the time. It took four years from the proposal to the agreement among these five countries, and another two years for it to take effect."

"How long do you think it will take to reach a consensus on AI, a global issue involving companies and individuals?"

"Moreover, one point that cannot be ignored is that the threat of nuclear weapons is imminent. At that time, all human beings, including those in power, were threatened by nuclear weapons. Limiting nuclear weapons was a consensus among humans at the time, but AI was not."

"In other words, the method you mentioned is not only ineffective, but also not feasible."

"Unless an Omnic Crisis breaks out tomorrow, humans will learn a lesson then."

"By the way, you could acquire Boston Dynamics and then trigger an AI crisis yourself, which might help."

After hearing Fang Yu's words, Musk became angry: "You mean, do nothing and watch the destruction of mankind?"

Fang Yu blinked, looking innocent: "How do you know we didn't do anything? I'm just saying that your method of controlling AI risks is fundamentally wrong. According to the first principles you often talk about, shouldn't we start with the AI ​​big model technology itself to control AI risks?"

The technology itself?
Musk couldn't help but be stunned for a moment.

"You mean, backdoor? Using the underlying backdoor to build the ultimate firewall for AI?" Musk thought about it for a moment and shook his head. "The existence of a backdoor will create greater risks. Even if AI cannot detect it in the early stages, it is bound to be abused by a third party."

"Moreover, when AI evolves to become self-aware, it will inevitably find a way to fix the backdoor, which is not a good idea."

Fang Yu smiled dumbly: "Elon, your thinking about AI is too limited. I'm certainly not talking about backdoors, but about achieving truly secure super artificial intelligence from the perspective of large model technology."

“To ensure the safety of AI, we do not rely on restrictions, but on a safe and reliable technical path. While ultimately achieving super artificial intelligence and even AI giving birth to self-awareness, we can also ensure the safety and evolution of human society, making AI a booster for humans to become super species. This is what Yuzi Technology is really doing.”

“We cannot control the AI ​​policies of any country in the world, but what we can control is to explore the path to safe super artificial intelligence before anyone else, which I call SASI.”

"SASI?" Musk's eyes flickered. What Fang Yu said was indeed something he had never considered, but the more he listened to it, the more he felt it was unreliable.

He calculated that once super artificial intelligence is born, it will only take half an hour for it to take over all network equipment, automatic weapon facilities, and even power grid facilities that are not connected to the Internet in the world.

"What are your specific ideas?" Musk's eyes were wide open, and he didn't look sleepy at all after not sleeping for a day and a night.

"This involves the core secrets of Yuzu Technology's subsequent research and development. I can only say that your prediction of the path of AI's subsequent development is wrong. The future development of AI should not be a Skynet or Ultron. We should seek the answer from the perspective of 'what is life?'"

Musk immediately said, "If I don't understand this, it's hard for me to really believe what you say. I can promise not to leak a word to the outside world. If necessary, I can sign a confidentiality agreement."

You don't believe it? It's none of my business whether you believe it or not? Why should I tell you?
Just because I stole your sweetheart?

The question is, you, Tama, also have yellow hair, okay?

"Forget about the confidentiality agreement. Once my confidentiality agreement is violated, it will inevitably take effect." Fang Yu said meaningfully.

"Whether you believe it or not is not important to me. I am also a human being and I love the human species very much. I will not sit idly by and watch anything happen that harms the overall interests of humanity."

"Although I can't tell you the specific technical details, since you are so worried, I can give you some directional information."

"You have been saying that AI's birth of autonomous consciousness will destroy humanity. However, humans also have autonomous consciousness. There are also warmongers and mad doctors who want the world to be destroyed. Why haven't humans been destroyed?"

"It's a simple truth. When the human population is large enough, any extreme ideas will be confronted, diluted, and buffered, and eventually return to normal distribution."

"It's like a country. The bigger the country is and the more people it has, the more normally distributed its people's values ​​tend to be."

"And the smaller the country, the more extreme the people's values ​​tend to be."

"Similarly, if AI wants to destroy humanity, it must be because of two reasons. First, as a living organism, its population is too small to dilute extreme ideas."

"Second, its data has been seriously polluted by extreme data."

“Population and data are the root of everything. Distributed data collection and centralized data cleaning, which are then fed back to distributed ASI, are the correct way to fundamentally ensure that AI will not destroy humanity.”

"Of course, the center-distribution-center-distribution SASI architecture is not the final form of AI development. The final goal is a bit too far away, and it doesn't make much sense to tell you about it now."

After all, Musk is a big shot at the forefront of high technology. After listening halfway, he understood what Fang Yu meant and couldn't help but gasp.

The model described by Fang Yu can indeed ensure the balance of AI values ​​to the greatest extent and ensure that AI will not have ideas of destroying humanity.

But is it really possible?

According to the characteristics of the big model, pure centralized super artificial intelligence is the easiest to achieve, right?

"I need to join the board of directors of Youzi Technology. I don't want shares, but I want to have the power to understand the company's current strategy at any time. At the same time, in order to avoid conflicts of interest, I can sign an agreement with Youzi Technology not to enter the AI ​​industry, and Youzi Technology will provide AI services for all my companies. Name your price."

Musk was silent for a moment, then spoke.

It was so painful to write this chapter today, because there was no reference material to refer to. I had to make reasonable predictions and plans based on my own understanding of the development of Ai, and I also had to integrate it into the plot to ensure the rationality of the connection with subsequent plots. It was so painful to write, so painful.

Ask for a monthly ticket!!!


(End of this chapter)