Technology invades the modern world

Chapter 390 The Victory or Defeat Is Not on the Chessboard

Lin Ran's live streams are not only watched by Chinese netizens, but also by netizens around the world.

Although Lin Ran was broadcasting live in Chinese, without real-time translation, overseas viewers couldn't immediately understand the content.

But there are reporters.

All major newspapers have branches in China, responsible for reporting news from China.

Where do the famous BBC and AFP filters come from? It's because they have personnel and business operations in China.

Chinese media outlets also have overseas propaganda tasks, and are responsible for publicizing events happening in China.

The left and right brain chip architecture is a good medium for promoting China's technological progress.

In addition, there are many self-media bloggers who will spread Lin Ran's live broadcast.

So the day after Lin Ran's live stream ended, his clips were translated into English and widely circulated around the world.

Those who pay the most attention to Lin Ran's views are his peers in Silicon Valley.

Europe is still working on how to formulate regulations, but France is full of ambition as they are about to launch their own artificial intelligence: Lucy.

It's precisely because of "Lucy" that France has recently become somewhat unenthusiastic about the Brussels conference on AI regulation. "Now that my Lucy is here, do you still want to regulate our AI? Absolutely not! We need to grant our 'Lucy' a special pardon. 'Lucy' is a treasure of all Europe!"
When the concept of left and right brain chips was proposed, colleagues in Silicon Valley were discussing it.

Besides considering the competition, they are also wondering if this will lead them to a new direction.

Sam Altman arrived at OpenAI's San Francisco headquarters on the second day of the new year.

With each round of funding for OpenAI, its valuation has reached record highs, and similarly, the pressure on CEO Zhang Yimou has been increasing.

They hope to push OpenAI's valuation in the primary market to $300 billion during the funding round in August.

What could be more effective at fooling investors and inflating the bubble than making empty promises?

The Netscape Navigator case is still fresh in the minds of people like Sam.

Perhaps left-brain/right-brain chips could be an opportunity, Sam Ultraman believes.

Some OpenAI executives came to the company for meetings, while others joined the discussion remotely via video conference from their vacation spots.

No one dared to underestimate Lin Ran.

The closer the competitor, the more importance is attached to them.

OpenAI has been researching DeepRed for a long time, and the more they research it, the more they admire its ability to create better commercial products with fewer graphics cards and lower computing power than OpenAI.

This is extremely difficult; they cannot do it, and they are eager to know how their Chinese counterparts have managed to do it.

Silicon Valley universities have sent many offers to graduates of Lin Ran's math class. If you write that you have interned at Crimson on your resume, these universities will light up like vultures seeing carrion and will immediately issue an offer. The application was written yesterday, and less than 24 hours had passed when the offer was sent to China.

After this phenomenon was posted on the Simplified Chinese internet, netizens joked that it was just one of many "Lin Ran phenomena," which further boosted Jiaotong University's admission scores.

The information Lin Ran revealed during the live stream seems to have given OpenAI researchers a glimpse into the tip of the iceberg of Crimson's secrets.

At the start of the meeting, CEO Sam Altman sat at one end of the table, and his secretary turned on the large screen, displaying an architecture diagram created by OpenAI researchers based on the left and right brain chip architecture proposed by Lin Ran.

The first to speak was Technical Director Chris Valho, who calmly analyzed the situation:

"Theoretically speaking, the idea of ​​the left and right hemispheres having different functions is undoubtedly reasonable."

Perception and reasoning tasks do have different computational characteristics, but I think Professor Lin may have overestimated the role of hardware in this division of labor.

Current deep learning frameworks can achieve parallel processing of tasks, and most efficient inference tasks can be completed on the same hardware.

Can different computing methods truly change the existing architecture? I'm skeptical.

Chris's question raised some questions, but he didn't give up easily.

No sooner had he finished speaking than Ilya Sutzkeffer, the chief architect in Tokyo, responded immediately: “You’re right, existing frameworks can handle the parallelization of some tasks, but this is only a partial optimization for perception tasks.”

However, we all know that existing hardware bottlenecks exist.

In the inference process of the GPT series models, a large number of inference calculations involve long inference chains and symbolic computations, which are not very efficient for current GPUs, especially as the model size continues to increase.

We can certainly learn from Professor Lin's concept and construct a dedicated logical reasoning processing unit, separate from the perceptual task processing unit.

Ilya paused, then added, "Moreover, we must note that IBM and Intel have been exploring this direction for many years."

As early as ten years ago, IBM proposed the concept of a multi-processing unit architecture, which aims to improve the efficiency of AI reasoning and perception by dividing computing tasks into different modules.

Intel, on the other hand, proposed the concept of neuromorphic computing chips in 18, attempting to mimic the functions of the brain and optimize the computational interaction between perception and reasoning tasks.

I think Professor Lin simply summarized Intel's approach in a more accessible and understandable way.

Professor Lin, building upon the work of IBM and Intel, theoretically argued that if we want to continue pushing artificial intelligence towards generalized AGI, we need and must take this path.

Researcher Andrew Kaplan said, "That's right. After watching Professor Lin's speech on artificial intelligence, I specifically looked up some information. Intel launched their first neuromorphic chip, Loihi, in 2017."

“Neuromimetic chip?” Sam Altman repeated the term.

Andrew nodded and said, "That's right. As we all know, the chips we use now only have 0 and 1. They use diodes to simulate the states of 0 and 1 to build a huge computing network. Its core is binary."

However, the human brain has more than two signals. Neurons are connected to each other through synapses, and a neuron can be connected to many neurons at the same time.

Loihi, on the other hand, is a chip that simulates neurons.

第一代的Loihi每60平方毫米的芯片就有13万个神经元，他们在2021年推出的第二代Loihi芯片，拥有128个神经形态核心，Loihi 2每个核心的神经元是第一代的8倍的神经元。

Meanwhile, each neuron in Loihi 2 can be assigned up to 4096 states according to the model, compared to only 24 in the first generation. The neuron model is similar to an FPGA, fully programmable, and has greater flexibility.

Simply put, Silicon Valley already has right-brain chips.

Intel's Loihi uses a neuromorphic computing architecture to try to mimic how the brain processes perceptual information, and proposes the concept of perception-reasoning fusion.

In practical applications, these neuromorphic chips can perform well in certain tasks, but their performance is not significant in complex reasoning tasks. Compared with traditional logic processing units, they do not achieve the expected performance improvement.

The problem is that the architecture proposed by Professor Lin requires us to utilize both right-brain and left-brain chips. The GPU for the left brain is already mature, and the Loihi for the right brain is also relatively mature.

We have ample resources to test this technical approach.

What we need to do now is combine the left and right hemispheres of the brain, Nvidia's left brain and Intel's right brain.

Amy Zhang, the head of the open-source team, said:
"I think this theory of left-brain and right-brain specialization is too idealistic, at least for now, its application value is not great. Can we have a clear case to show a scenario where perceptual tasks and reasoning tasks must be processed separately? At present, most tasks are highly parallel tasks, and optimization on the same hardware can achieve sufficiently good performance."

Are we going to rebuild the entire hardware architecture?

Ilya responded, "If we just strengthen the existing hardware architecture, we will remain dependent on massively parallel computing."

The real breakthrough lies in the division of labor between the left and right hemispheres of the brain and the physical isolation of different computational tasks, thereby achieving more efficient energy utilization and computational optimization.

Furthermore, did you listen carefully to Professor Lin's speech? He revealed a secret in his speech, namely the corpus callosum.

This is a structure in the human brain, and I believe the secret to combining left and right brain chips lies in the corpus callosum.

Professor Lin works for his country and serves its interests, but I believe he possesses a spirit that resonates with all humanity. He is using this to suggest that the corpus callosum is a concrete key, the key to unlocking the specific design and structure of the left and right hemispheres of the brain.

Lin Ran found what he heard baffling. "You guys have quite the imagination. I was just giving an example. How could you come up with so much?"
Andrew Kaplan added: "As Professor Lin said, we have clearly felt that its ceiling is there."

Without changing the hardware architecture, we will never be able to break through the ceiling.

Changing the hardware architecture will bring us entirely new possibilities.

Furthermore, Intel is an American company, which gives us a natural advantage in this regard.

"But this approach also presents enormous challenges," Chief AI Researcher Jeffrey Hopkins suddenly interjected. "Reconstructing the entire hardware architecture would involve costs and time far exceeding our current expectations."

Furthermore, we need to consider whether the collaborative mechanism of left-brain and right-brain chips can fully reflect the efficiency of real-time data exchange, and how to solve the data exchange bottleneck.

Most importantly, how can we be sure that what Professor Lin said is true?

Could this be a conspiracy, a scheme to divert the competitors' attention and energy?

Ilya said, "Professor Lin mentioned in the live broadcast that the relevant theoretical verification papers will be published, and he will demonstrate the necessity of the left and right brain architecture."

Jeffrey Hopkins corrected him: "That's right, but the problem is that just because the theory is true doesn't mean it can be put into practice in engineering, and it doesn't mean that current hardware technology is sufficient to support its implementation in engineering."

Just like President Reagan's Strategic Defense Initiative (SDI) in the 80s, does it make sense? Yes, it does. If I intercept all the missiles, wouldn't the Soviet Union's nuclear weapons be unable to reach me? Is it theoretically feasible? Yes, it is, as long as my guidance system is accurate enough, we can do it.

Wouldn't that eliminate any threat to us from Soviet Russia?

The problem is that even after spending hundreds of billions of dollars, it still hasn't been achieved, and it's still a long way from truly comprehensive interception.

The gap between theoretical feasibility and engineering practical feasibility

Hopkins paused for a moment, then stretched out his hands as if to embrace: "It may be bigger than the entire universe."

Amy added, "Yes, that's important."

This requires a highly efficient interaction mechanism. Without such a mechanism, two clearly defined hardware units might become two isolated systems, which would not only be inefficient but could also lead to more problems.

From a holistic perspective, Professor Lin's proposal is undoubtedly a forward-looking strategic direction.

The future of AGI will require various types of computing to work together, and we cannot rely solely on the existing unified architecture to support more complex tasks.

Suppose that our future models will not only enhance perception tasks, but also reasoning, planning, and emotion understanding, which have completely different computational characteristics.

If we can't divide the work and cooperate at the hardware level, these tasks will slow down overall efficiency, which would be counterproductive.

Jeffrey Hopkins said, "I also believe this is not a passing fad, but part of Crimson's long-term strategy."

We must also focus on the future intelligent architecture, taking into account the deep integration of hardware, algorithms and applications, in order to take a leading position in the next generation of artificial intelligence breakthroughs.

The problem is, when should we start? How much effort will it take? With limited manpower, how can we achieve this kind of architectural integration?
These are all difficult decisions to make.

In the past, China has always followed us, and this time we can follow China again. We can catch up after China achieves relatively good results on this framework.

Sam Altman was already impressed after listening to the whole thing. He wasn't an AI professional; he was an angel investor, an investor. He didn't understand technology, but he understood business.

He had already heard what he wanted from the experts' discussions.

That's what we call expectations; to put it nicely, it's called expectations; to put it bluntly, it's called hype.

Sam Altman waved his hand: "That's not a problem. Intel's Loihi chip is still a complete failure to this day. Apart from its scientific research significance, it has not played its due role in the actual market at all."

All we have to do is propose a partnership with Intel, and Intel will cover the costs, personnel, and computing cards for us.

Nvidia dominates the left brain of the chip market, with a market value of four trillion US dollars, while Intel's market value is only one hundred billion US dollars.

"A forty-fold difference—if they can occupy the niche of right-brain chips, they can reach Nvidia's position."

The other researchers immediately realized, "That's right, we can get something for nothing!"

Andrew Kaplan pointed out, "That's right! Not just Intel, but IBM too."

IBM's TrueNorth chip also attempts to simulate the structure of the brain's neural network, while Intel's Loihi has shown excellent parallelism in perception tasks. Their hardware faces the same problem: when the task becomes more complex, the coordination and division of labor between hardware becomes particularly important.

We could use this as leverage to test the sincerity of Intel and IBM.

As Professor Lin said during the live broadcast, he submitted his requirements to four chip companies in China, and only the winner could become his partner.

We can also use the same method to make Intel compete with IBM.

With both sides bidding, the price will be high enough. Intel and IBM are both symbols of America Technology and are genuine American companies, which means they can definitely offer a particularly high price, Sam Altman thought.

New Year's Day in 2025 falls on a Wednesday, so the following day is a workday, which is also the day the stock market opens.

Early in the morning, China Mobile was the first to jump the gun, its stock price soaring straight up and hitting the daily limit.

With a market value exceeding 2 trillion RMB, China Mobile's move can definitely be described as a giant dancing, and it has also triggered a surge in the entire state-owned enterprise sector.

"Even the god of fire said he uses China Mobile! How could China Mobile be bad? It's definitely a national brand, and its current valuation is still too low! Let's set a target: four trillion!"

"Burning God's mobile number is from China Mobile. What does that mean? It means that China Mobile enjoys high official trust, and that it is highly reliable. Political reliability is the greatest reliability! What reason do we have not to buy from such a reliable company?"

On Eastmoney's stock forum, the guys there are no different from those on Baidu Tieba; they all look for reasons behind the results with unwavering conviction.

Even if there's no result, they can still imagine one and say that their stocks will definitely rise, there's no reason why they shouldn't.

Every time the market drops, it's because the big players are manipulating things against me. They say things like AI is monitoring my account, and every time I buy, the price drops and every time I sell, the price rises. They say the entire A-share market is working together against me.

The chip sector also surged, with Cambricon leading the gains, hitting the daily limit.

"The concept of right-brain chips alone is worth $4 trillion! Nvidia has a market value of $4 trillion for left-brain chips, while Cambricon, a right-brain chip company, has a market value of less than 300 billion RMB. One limit-up? At least 10 limit-ups to reach 2 trillion RMB first."

Similar posts are countless in Cambricon's stock forum. One retail investor (a "leek") timidly asked, "Didn't you say there would be four companies in total, and that we'd be competing with Huawei? How come you're making it sound so certain?"

"? You wouldn't even dare to think about that, let alone speculate on stocks! Stock trading is all about expectations; buy when there's disagreement and sell when there's consensus! It's precisely because of competition and disagreement that there are buying opportunities, and that's when funds flow in!"

If it's Cambricon, Cambricon has already been selected, so how could it be your turn to buy it?
Buy where no one cares, sell when everyone is talking!
Institutions and speculative funds that received the news started buying up the dip ages ago, okay?

"Where did this newbie come from? Don't you even know this?!"

"No, this is also a case of nobody being interested? Cambricon is almost hyped up. It was hyped up once before, and its price-to-earnings ratio was sky-high. You're telling me nobody's interested?"

"With four companies plus Huawei, it's like choosing one out of five. As the only listed company, doesn't that mean Cambricon has the best chance? However, I still hope to hear news that DeepRed Technology has invested in Cambricon."

Huawei will never let you invest, but Cambricon will. I guarantee that everyone at Cambricon, from management to shareholders, will wholeheartedly support it. DeepRed Technology will invest.

Once you invest, you can secure a place in the right-brain chip industry by virtue of your equity relationship.

I urge the board secretary of Cambricon to quickly mobilize all available resources! Now is the time to act! As a retail investor, I'm getting anxious!

The stock forum is buzzing with discussion, with everyone advocating for their own understanding and their own accounts.

On the first working day of the new year, Cambricon's board secretary received 800 phone calls.

A retail investor called in, insisting that he be the one ultimately selected as the company.

Organizations from Beijing to Shanghai, from the Yangtze River Delta to the Pearl River Delta, want to come to their company for research.

They worked non-stop all day.

As for the other three companies, the investors went directly to the companies, demanding an explanation.

Moore Threads is not yet listed, and none of these three companies are listed. In 2022, Moore Threads signed a strategic cooperation agreement with Zhejiang Wasu, a subsidiary of Wasu Media. Although there was no specific business cooperation between the two, Wasu Media's stock price hit the daily limit upon opening.

H&T, which holds a 1.244% stake in Moore Threads, also hit its daily limit.

Any company that has even the slightest connection to a related company sees its stock price soar.

"No, Mr. Zhang, why weren't we informed of such important news before?"

The conference room at Moore Threads was packed with representatives from various investment institutions.

Like Cambricon, Moore Threads is also a domestic GPU company. Unlike Cambricon, which focuses on AI chips, Moore Threads makes chips for both B2B and B2C markets, meaning they also make gaming graphics cards for ordinary gamers.

Its founder, Zhang Jianzhong, was formerly the CEO of NVIDIA Greater China and the Global Vice President. Because his name contains the character "忠" (zhong, meaning loyalty) like Zhang Zhongmou's, this resume and this coincidence have made many investors very optimistic about its development.

This is an even bigger surprise.

Unexpectedly, there was an unexpected gain.

The company I invested in actually became a partner of DeepRed, and not just one that's superficially related, but one that deals with right-brain chips—such a core concept.

You can't say the brain isn't important, can you?

As for whether they will have to compete with Huawei in a quarter of the time, that's not important. In the eyes of these investment firm executives, it's as if they've already been selected.

It's all up to human effort. There are only four competitors left. Is it really that difficult to stand out from the crowd?
In some ways, these executives think no differently from retail investors.

“It’s not that I don’t want to tell you, but we only received the news recently, around November of last year, when Crimson and we had our first round of contact,” Zhang Jianzhong said with a wry smile.

Those present were all major investors, and no one dared to offend them, including Shenzhen Capital Group, Xiaoma Executive, Sequoia Capital China, Shenhai Guosheng, CCB International, Qianhai Fund of Funds, and so on.

State-owned capital is the main source, with some foreign and private capital.

"Then we had many rounds of contact with DeepRed, and finally they confirmed that we were qualified to participate in their program competition. We are currently in the program design stage, and we need to have a program and a demo before they will determine whether we can be selected based on our chip design."

The whole process will take at least another year; it won't be that quick.

I know everyone is excited. I was just as excited as you when I found out. This is extremely important to us at Moore Threads. It even relates to whether we are a mediocre company or a great company.

"We are all putting in 200% effort in our preparations," Zhang Jianzhong said earnestly.

The atmosphere in the meeting room was lively, just like the atmosphere at the A-share market today.

"Efforts cannot be just tactical efforts; they must also be strategic efforts!" Chen Hongyi, a senior executive at Shenzhen Capital Group, said with an equally serious attitude: "There are so many prominent figures in the investment circle here."

Mr. Zhang, if you had told us sooner, we could have started working sooner.

I understand your thinking; you want to manage expectations in advance, fearing that if we have too high expectations, it might cause a negative chain reaction if we are not selected.

You're thinking about the left and right brain structures, right brain chips—this must be highly confidential. Given the confidentiality requirements of Shenhong and General Manager Lin, and the recent incident at Huawei, even if it's not ultimately selected, the outside world won't know about this process.

Unexpectedly, Mr. Lin himself revealed it during the live broadcast.

No, don't you think this can be kept secret? Are these the same thing?
It involves four companies, including Huawei. How could you possibly keep it a secret?

Other organizations have already taken action, shouldn't we take action too?
Mr. Zhang, you have a technical background, and we respect your expertise and have confidence in Moore Threads' R&D capabilities. The problem is, often, victory or defeat lies outside the chessboard.

Your confidence is a good thing, but you can't let it jeopardize the future of the entire company!

Chen Hongyi spoke with varying degrees of emphasis, but his dissatisfaction was evident in his words.

Zhang Jianzhong said with a wry smile, "We think that technology should be the key. We have already poached two engineers from Intel and IBM who have worked on neuron chips that are somewhat similar to right-brain chips. At the same time, we are also looking for engineers in the market whose direction is compatible."

Chen Hongyi sighed and said, "President Zhang, this is not just a technical matter."

Didn't you understand?
The outcome is not determined on the chessboard.

Biren Technology is backed by Tencent, and I don't need to elaborate on the relationship between Tencent and Shenhong, right?
To put it bluntly, the four GPU giants that Mr. Lin mentioned yesterday are all doing poorly right now, and they are all losing money every year.

However, once they become suppliers of right-brain chips, Crimson and Apollo Technologies have virtually unlimited resources, and even a small order can turn any of them into a profitable company.

Not to mention, this means that it will have a place in the future AI ecosystem.

The interests involved go beyond what technology can determine.

I'm pretty sure Biren Technology must have spoken to Tencent and asked them to do their bidding.

Furthermore, Moore Threads is a company based in Beijing, while Biren Technology is a local company in Shenhai. Do you think the local government in Shenhai would prefer you or Biren Technology to be elected?
Will Shenhai intervene in this matter?
Bi Ren is definitely trying, to put it more bluntly, they are absolutely trying to become a supplier of right-brain chips and gain the support of the Crimson, just like TSMC supported ASML and OpenAI cooperated with Nvidia in the past.

You're already three months behind, but thankfully three months isn't a long time. We still have time to catch up.

Mr. Lin mentioned during the live broadcast that four companies are competing, which means that Biren Technology has not yet achieved final victory.

That's good, that's good.

Zhang Jianzhong even sensed a feeling of relief and joy on Chen Hongyi's face; it was the first time he had ever seen such an emotion in this financial executive.

Zhang Jianzhong said, "President Chen, I believe President Lin will give us a fair chance to compete. I think we are a technology company, and ultimately we should rely on technology to speak for ourselves."

From top to bottom, the entire army fought with all its might, determined to win.

Using underhanded tactics to try to gain ultimate victory through shared interests is certainly a business-like battlefield, and no one will condemn the victor for moral reasons—I understand that principle well. But this time is different.

This time, we are competing for the key to the future. I have a feeling that left and right brain chips will become the key for humanity to access universal AGI, and we are competing to become one of the materials for forging this key.

This time, I'm determined to win through pure skill, not through anything outside the game of chess.

(End of this chapter)