Black technology: a super empire spanning two worlds

Chapter 370 [The sooner the better, a lightning-fast slap in the face]

Upon learning of Xiao Yu's intentions, Xu Changming was surprised but also invigorated, asking, "How exactly do you intend to undermine him?"

At this moment, Xiao Yu slowly said, "This ancient civilization of five thousand years has taught us a very important lesson: those who play dirty tricks can never beat those who play open tricks. I plan to use an open trick, and announce that the new version of AI Xiaojing will be open source on the same day it is released."

Upon hearing this, Xu Changming was stunned. Xiao Yu calmly said, "In the future, the value of an open-source AI Xiaojing ecosystem will far exceed that of a money-making closed-source OpenAI and Cyera-AI."

"This open-source move is intended to shatter the existing valuation logic of the AI ​​ecosystem of the US-based company, making it impossible for the entire AI ecosystem of the US-based company to raise funds. Not only will the money that was burned so wildly in the past have no hope of being recouped, but the company will also become increasingly hesitant to invest in the future. I want to use the open-source of AI Xiaojing to tell US investors that the so-called hardware moat and computing power moat are all nonsense."

Xu Changming was shocked: "This move is a bit ruthless. Cyera-AI is currently valued at $5 trillion, Nvidia at $4 trillion, Microsoft at $4 trillion, and other AI companies have valuations that are higher than the last. If the valuation logic of the AI ​​ecosystem is no longer supported, I can't imagine how these companies' valuations will be slashed."

Especially considering Cyera-AI's current market capitalization of $5.32 trillion, if it were to collapse, it's unimaginable how badly the capital market would crumble.

Xiao Yu said methodically, "After announcing the open source, we must also promise that AI Xiaojing will abide by the MIT license. At that time, it will be the only top-level AI agent model in the world that abides by the license, allowing anyone to use, modify, and distribute it at will. Any results made based on the AI ​​Xiaojing agent model will not require Lingjing Technology to receive a single penny, nor will it require Lingjing Technology's consent."

Upon hearing this, Xu Changming gained a deeper understanding of the phrase "drawing the firewood from under the cauldron" that the big boss had mentioned.

Without a doubt, once this happens, the world's top universities and leading companies will absolutely choose to get on board. And they won't be able to compete if their peers do.

In this way, whenever AI Xiaojing releases a new version, everyone who has already joined will have to follow Lingjing Technology's lead.

The problem is, they never know when the next version will come, or what it will be like.

As a result, every time AI Xiaojing releases a new version, all their AI application software and even hardware chips have to adapt to it in order to stay ahead of the competition, especially their peers.

Even if OpenAI and Cyera-AI are forced to open source by then, it will be useless, as others will not easily migrate, because they have already invested huge sunk costs in the AI ​​Xiaojing ecosystem, which means that the cost of changing platforms will be very high.

Moreover, if OpenAI and Cyera-AI are not as good as AI Xiaojing, then it is even less likely that they will be transferred.

"I finished the major version 3.0 of AI Mirror last year, but the timing wasn't right yet. I was waiting for Lao Mei to make his move," Xiao Yu said with a smile.

The concept of Stargate can be traced back to the time of Amy Oldden. Last year, OpenAI, Microsoft, and Cyera-AI were planning to build a data center called "Stargate," but it was unable to proceed due to regulatory constraints.

After taking office this year, Deng Xiaoping revoked the AI ​​executive order signed by Deng Xiaoping on the very first day of his term. Instead, he accelerated the implementation of projects through a deregulation strategy and announced the "Stargate" plan on the same day.

Xiao Yu said in a deep voice, "Now is the time. Since Lao Mei's 'Stargate' project has been announced in a high-profile manner, we should not delay and should directly slap him in the face at lightning speed. The 3.0 version will be launched tomorrow, and we will also announce that it is open source."

Upon hearing this, Xu Changming chuckled and said, "I reckon the most embarrassed person in the global AI community will be Lao Sun. He's poured a fortune into 'Stargate,' and with this move of ours, Lao Sun will be in a terrible predicament."

That's right. If we keep investing money, it's like handing someone a free ride to be fleeced. If we don't invest any money, Dongwang will probably fleece him first.

……

The following day, Wednesday, January 22nd. Just as the global AI community was focused on the "Stargate" issue, Lingjing Technology released a major announcement.

AI Mirror 3.0 has been officially launched. It's the first major version release in nearly two years, but that's secondary. The other two pieces of news are the real bombshells, causing a sensation in the global AI community.

Along with launching its 3.0 major version, Lingjing Technology announced that its AI Xiaojing 3.0 intelligent agent model is open source and has committed to adopting the MIT license. This makes Xiaojing 3.0 the only top-tier AI intelligent agent model in the world that complies with the MIT license.

Although industry insiders generally believe that Cyera-AI has caught up and surpassed Lingjing Technology in the field of AI, AI Xiaojing is not something that other competitors can compete with and still belongs to the top tier.

This new version 3.0 is not only an epic iteration update, but it also directly announces that it is open source and complies with the MIT license, allowing anyone or any organization to use, modify, and distribute it at will, without paying a penny to Lingjing Technology or obtaining Lingjing Technology's consent.

This news caused a complete uproar in the AI ​​community, with many finding it unbelievable and even initially believing it to be fake news. Another piece of news was equally explosive: Xiao Yu's latest update on his personal Weibo account.
More powerful capabilities, lower energy consumption, and a lower barrier to entry – this is the latest major version of our Little Mirror 3.0.

Our technical team bypassed NVIDIA CUDA in version 3.0 by using a lower-level programming language for optimization. This means that the agent model in version 3.0 is capable of being optimized at the assembly level using other GPUs, thus freeing it from dependence on CUDA and NVIDIA.

One of the biggest highlights of version 3.0 is the exponential reduction in model training costs.

In our internal testing, the deployment of version 3.0 of Xiaojing only required 1024 GPUs for one month of training, costing 50 RMB, but its performance was already comparable to Cyera-AI's most advanced version. Simply put, its cost was only 0.75% of that of its competitors.

Most importantly, all of this is open source, and any individual or organization can use it freely and for free without needing Lingjing Technology's permission.

Version 3.0 is more powerful than you can imagine. I can't summarize it all on Weibo. You'll have to try it out for yourselves to see, especially since it's open source and free.

……

Whether it's open source or bypassing CUDA, it's an absolutely bombshell event for the AI ​​community.

CUDA is a high-level GPU programming language developed by NVIDIA, similar to C++ and Java used by CPUs. Its advantage is that its semantics and syntax are closer to human language, but its disadvantage is that it results in a loss of runtime efficiency.

For AI Mirror 3.0, Xiao Yu chose to bypass CUDA and adopted PTX, a lower-level language closer to the machine, and thoroughly optimized the programming at the PTX level.

The previous PTX language had the advantage of high execution efficiency, but it was not very user-friendly for human understanding. Therefore, programmers used CUDA programming, which is one of NVIDIA's most important competitive advantages.

Xiao Yu solved the pain points of the PTX language, directly eliminating the dependence on CUDA and NVIDIA, allowing the AI ​​Mirror 3.0 model to effectively utilize any GPU.

Amex has even banned the sale of the crippled version of the H20 chip, so it doesn't matter now, since there are other cards on the market besides Nvidia cards.

Lingjing Technology suddenly unleashed such a bombshell at this time, leaving the entire AI community sleepless all night.

……

(End of this chapter)