Nearly thirty years old, established in science and technology

Chapter 75 Microsoft’s life-saving magic weapon (please vote for me)
A depth bomb that subverted the cognition of countless AI people was thrown into the Internet, and many people almost lost their minds after being bombed!

Typical representatives are NVIDIA and Microsoft.

Since Huang Renxun returned to China last time, Microsoft OpenAI and NVIDIA quickly reached an agreement on hardware procurement.

OpenAI even announced in a high-profile manner that its fifth-generation GPT model already has powerful [human-like generalization capabilities], and the second-generation smart Windows in cooperation with Microsoft has officially launched.

What is human-like generalization ability?
In layman's terms, it is the power of association and the ability to draw inferences from one instance.

For example, in some forums, people often say "Qin Shihuang touched the wire - he won!" 】.

As a person, if you know the surname of Qin Shihuang and what it feels like to touch a wire, you can understand that this sentence of the netizen is talking about Win!

But for machines, it's hard.

This ability is [Human-like Generalization Ability]. Fundamentally speaking, it is almost the same concept as Juan'er Intelligent Technology's [Human-like Thinking], but it is one level weaker and is still achieved by accumulating computing power - among other things The essence is a systematic reasoning that combines vocabulary.

But for Microsoft, whether it can do it or not, this is the essential difference. As for whether to accumulate computing power, it is nothing more than a matter of money. Is Microsoft short of money?
Therefore, they are working hard internally to develop the next generation of Windows, determined to replicate all the functions currently implemented by Juan Er.

This is Microsoft's life-saving weapon.

But today...

"Quickly, go back, yes, this is the time!"

Qiao Beifeng, head of Microsoft's Widows business, and Altman, founder of OpenAI, were watching the short video sent by Hezhou University frame by frame. Even at critical moments, they had to go back and watch it again.

"How is it? Real or fake?" Qiao Beifeng was not professional about this, so he asked Altman.

"I don't know. This is the first time I've seen someone train AI like this."

As the founder of OpenAI, he has been adhering to continuous exploration and innovation, even regardless of the cost, directly leading the trend in the ToC (consumer-oriented) field-until the emergence of curl-ear smart technology.

But at this moment, his eyes showed great disappointment - this kind of training method was completely beyond his understanding.

"How much better than our latest achievement?" Qiao Beifeng didn't ask whether it was strong or not. This was completely obvious!
"You mean MLC? That human-like generalization ability neural network training method?" Altman shook his head, then looked up to the sky and sighed: "Just now, I even thought I saw God. Do you believe it, Qiao Bei?"

"Think!" Qiao Beifeng's heart thumped, and the hidden secret was broken - this person who has been completely arrogant since his debut seems to be on the verge of collapse - this is the collapse of his faith!

"Our method isn't bad, right? At least it's improved compared to before."

Qiao Beifeng is right. The new training method can save 40% of computing power in scenarios such as reasoning, understanding, and relationship processing, and the effect is better than before.

They have even found ways to imitate the intelligence capabilities of the Whispering system - even if the cost is indeed relatively high, at least it can be done.

"We all estimate that the set of equipment used to train Whisper with Curl's Smart Technology cost hundreds of millions. This is because they can only spend so much money."

Altman continued: "After watching this video, do you know what conclusion I have?"

"What conclusion?" Qiao Beifeng was confused, why did it involve money again? Didn't Nadella promise that there would be no upper limit on our costs?

"Whispering that thing probably cost nothing." After saying that, Altman lay down again, completely feeling like he had no love in his life.

"You didn't spend any money?" Qiao Beifeng's eyes widened. Are the rumors true?

The Whisper system was just a system that Ji Hong came up with to reduce his wife's workload. In terms of money, it only cost a computer equipped with a Moore Technology graphics card. "It's most likely like this." Altman leaned out from his chair and said, "Calculate the cost ratio. His computer costs 10,000 yuan."

"us-dollar? U.S. dollar?"

"No! It's Chinese Yuan, RMB!" Altman collapsed from his chair like crazy, and stretched out a finger to point around:
"How do we compare? We have prepared 400 billion US dollars, or billion yuan, to do this! They can do it with only ten thousand! Ten thousand! How many times, you can count how many times!"

Qiao Beifeng was also fascinated by Altman for a moment. How many times more powerful is this? Thirty thousand times?
"Thirty million times!" Altman was not confused yet, at least the numbers were calculated very accurately.

"Thirty million times, my God!" Qiao Beifeng was also frightened.

"So how do we compare?" Altman finally calmed down and lay down on the chair before:

"Don't say that our training method is ineffective. Even if it is effective, so what? We have finished today, and the curling ear intelligent technology will be able to train something more powerful and even beyond our understanding in minutes?
"How can we compete! Tell me how?"

"Calm down, calm down, calm down!" Qiao Beifeng stretched out his hands and looked at Altman who was about to jump up again and pressed his hands down: "I will contact the boss immediately."

Qiao Beifeng felt that he couldn't bear it anymore. This guy might go crazy and he had to say hello to the boss in advance.

"What happened?" Nadella listened to Qiao Beifeng's rants and finally understood what was going on: "Wait for me and go back right away!"

……

Microsoft is not at peace here, and the domestic network is also in trouble.

After the video was released, everyone, both professional and non-professional, watched it countless times.

This kind of training method has never been seen by the founders of OpenAI, let alone non-professionals.

"This really doesn't feel like training AI, it really feels like teaching students." Someone pointedly commented on Ji Hong's method.

In the past, training required a huge amount of data to support it, so it required powerful computing power accumulation support and a long time.

However, the data collected by Ji Hong today is not intuitively reflected in the video, but judging from the transmission time, the data cannot reach the GB level at a large scale.

For training AI models, how can this be considered large-scale data? It is simply small data!
"Teaching students? The students' understanding ability may be good, but are their analytical abilities that strong?" Someone immediately retorted:
"I think that this thing of Juan'er Intelligent Technology is more like a synthesis of humans and AI. It has human-like thinking and the super computing and analysis capabilities of machines. The two are combined."

"I don't understand what you are talking about, but let me tell you, it's true that the curling ear smart technology is awesome, right?"

"+1+1+1, seeing the two big guys chatting here for so long, I didn't even bother to talk, and finally caught the same guy as me, awesome!"

"+1"

……

In an instant, a group of "+1" followed behind. It seemed that there were quite a lot of people watching.

　Asking for a monthly ticket!

　　Eight thousand words of the third update have been sent out, please save one for posting later in the evening. Today’s total is 40,000 words.

　　My update rhythm has always been in the evening, and I am not used to posting in the early morning. It is too late for you book friends to read. The subsequent fixed update time is between 18:00 and 22:. I will try my best to keep the second place and the third place!

　　Finally, if you like the book, please remember to subscribe! Thanks!
　　
　
(End of this chapter)