Chapter 390: Victory And Defeat Not On The Chessboard
Lin Ran’s live broadcast was not only watched by netizens in China, but also followed by netizens globally.
Despite Lin Ran live streaming in Chinese language without real-time translation, foreign netizens couldn’t know the content right away.
But there are reporters.
All mainstream newspapers have branches in China, responsible for reporting China news.
How did the famous BBC filter and Agence France-Presse filter come about? Isn’t it because they have personnel and business operations in China?
Including China media, which also has the task of overseas promotion, responsible for disseminating what happens in China.
Something like the left-right brain chip architecture is a very good promotional medium that embodies China’s technological progress.
Additionally, there are plenty of self-media bloggers who will disseminate Lin Ran’s live broadcast.
So the day after Lin Ran’s live broadcast ended, his clips were translated into English and widely circulated globally.
The ones most concerned with Lin Ran’s viewpoint are peers in Silicon Valley.
Europe is still researching how to formulate regulations, but France is full of ambition, as they are about to launch their own artificial intelligence: Lucy.
And precisely because of “Lucy”, France recently has little interest in the Brussels meeting on artificial intelligence regulation. After my Lucy comes out, do you still want to regulate our AI? Absolutely not, we need to issue a special exemption for our “Lucy”, “Lucy” is Europe’s treasure!
After the concept of left-right brain chip was proposed, peers in Silicon Valley were all discussing this matter.
Besides considerations of competition, they are also thinking whether this will find a new direction for them.
OpenAI’s headquarters in San Francisco, on the second day of the new year, Sam Altman had already arrived at the company.
Every financing round of OpenAI creates history in valuation, and similarly, the pressure on him as CEO is growing greater.
They hope in the August financing process to push OpenAI’s primary market valuation to 300 billion US dollars at once.
What is better than painting a big pie to fool investors and continue blowing the bubble?
The Netscape browser case is still fresh in the memory for people like Sam.
Perhaps the left-right brain chip will be an opportunity, Sam Altman thinks so.
OpenAI executives, some came to the company for a meeting, others joined the discussion remotely via video conference from vacation spots.
No one dares to disregard Lin Ran.
The more they are peers, the more they value it.
OpenAI has been researching Crimson Technology, and the more they research, the more they admire how the other side makes better commercial products with fewer graphics cards and lower computing power than OpenAI.
This is very difficult, they can’t do it, they really want to know how peers in China achieved it.
Universities in Silicon Valley have sent plenty of offers to graduates from Lin Ran’s mathematics classes. If your resume mentions even one internship at Crimson Technology, these universities light up like vultures seeing carrion, directly issuing offers. The application was submitted yesterday, and the offer arrived in less than 24 hours China time.
After such phenomena were posted on the Chinese Internet, netizens joked that this is one of the many Lin Ran phenomena, which further boosted Jiaotong University’s admission score line.
And this time, the news revealed by Lin Ran in the live broadcast seemed to let OpenAI researchers glimpse the tip of the iceberg of Crimson Technology’s secrets.
The meeting began, CEO Sam Altman sat at one end of the table, the secretary turned on the big screen, displaying the architecture diagram made by OpenAI researchers based on the left-right brain chip architecture proposed by Lin Ran.
The first to speak was technical director Chris Volho, who calmly analyzed:
“Theoretically, the idea of left-right brain division undoubtedly makes some sense.
Perception tasks and reasoning tasks do have different computational characteristics, but I think Professor Lin might be overestimating the role of hardware in this division?
Current deep learning frameworks can achieve parallel processing of tasks, and most efficient reasoning tasks can also be completed on the same hardware.
Can different calculation methods really change the existing architecture? I am skeptical about this.”
Chris’s question raised some doubts, but they didn’t give up easily.
Just after he finished speaking, chief architect Ilya Sutskever in Tokyo immediately responded: “You’re right, existing frameworks can handle parallelization of certain tasks, but this is only partial optimization for perception tasks.
But we all know that existing hardware bottlenecks exist.
In the reasoning process of the GPT series models, a large amount of reasoning computation involves long reasoning chains and symbolic computation, and current GPUs actually aren’t efficient at handling it, especially as model scale keeps increasing.
We can completely borrow Professor Lin’s concept to build a dedicated logical reasoning processing unit, separate from the perception 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 multi-processing unit architecture, aiming to improve AI’s reasoning and perception efficiency by dividing computational tasks into different modules.
Intel then proposed the concept of neuromorphic chip in 2018, attempting to imitate brain functions and optimize computational interaction between perception and reasoning tasks.
I think Professor Lin just summarized Intel’s approach in a more accessible way.
Professor Lin stands on the foundation of IBM and Intel, demonstrating theoretically that to push artificial intelligence further toward general AGI, we need and must take this path.”
Researcher Andrew Kaplan said: “Exactly, after watching Professor Lin’s speech on artificial intelligence, I specifically looked up information. Intel launched their first neuromorphic chip Loihi in 2017.”
“Neuromorphic chip?” Sam Altman repeated the term.
Andrew nodded: “Yes, as we all know, the chips we use now only have 0 and 1, simulating 0 and 1 states through diodes to build massive computational networks, with binary at its core.
But the human brain has more than two signals; neurons connect via synapses, and one neuron can connect to many others simultaneously.
Loihi is a chip that simulates neurons.
The first-generation Loihi has 130,000 neurons per 60 square millimeters chip. Their second-generation Loihi chip launched in 2021 has 128 neuromorphic cores, with Loihi 2 having 8 times the neurons per core as the first generation.
At the same time, Loihi 2 allocates up to 4096 states per neuron based on the model, while the first generation was limited to 24; the neuron model is similar to FPGA, fully programmable, with greater flexibility.
Simply put, Silicon Valley already has the right brain chip.
Intel’s Loihi adopts neuromorphic computing architecture, attempting to imitate the brain’s way of processing perception information, proposing the concept of perception-reasoning integration.
In actual applications, these neuromorphic chips can leverage advantages in certain tasks, but in complex reasoning tasks, the effect is not significant, and compared to traditional logic processing units, they haven’t achieved the expected performance improvement.
But the problem is, with the architecture Professor Lin proposed now, we need to utilize both right brain and left brain chips simultaneously. The left brain GPU is already mature, and the right brain Loihi is relatively mature too.
We have sufficient conditions to test this technical route.
What we need to do now is to combine left and right brains, Nvidia’s left brain and Intel’s right brain.”
Open source team leader Amy Zhang said:
“I think this left-right brain division theory is overly idealized, at least its application value is not significant currently. Can we have a clear case showing scenarios where perception tasks and reasoning tasks must be handled separately? Current stage tasks are almost all highly parallel tasks, where optimization on the same hardware can achieve sufficient performance.
Are we going to rebuild the entire hardware architecture?”
Ilya responded: “If we just reinforce the existing hardware architecture, we will always be stuck relying on large-scale parallel computing.
The real breakthrough is through left-right brain division, physically isolating different computational tasks, thereby achieving more efficient energy utilization and computational optimization.
And have you all seriously listened to Professor Lin’s speech? He revealed a secret in it, that is the corpus callosum.
This is a structure in the human brain. I think the secret of combining left-right brain chips lies in this corpus callosum structure.
Professor Lin works for his country, serves the interests of his country, but I believe he has the spirit of all humanity, hinting to us through this way that the corpus callosum is the specific key, the key to unlocking the specific design structure of left-right brains.”
If Lin Ran heard this, he would feel baffled. Your imagination is really strong. I was just making an analogy, and you thought of all this?
Andrew Kaplan added: “Just as Professor Lin said, we have clearly felt that its ceiling is right there.
Without changing the hardware architecture, we will never break through the ceiling.
Changing the hardware architecture will bring us entirely new possibilities.
Moreover, Intel is an America company, we naturally have an advantage in this aspect.”
“But the challenges this approach brings are also enormous.” Chief AI researcher Jeffrey Hopkins suddenly interjected: “Reconstructing the entire hardware architecture involves costs and time investment far beyond our current expectations.
Furthermore, we need to consider whether the synergy mechanism of left-right brain chips can fully realize real-time data exchange efficiency? How to solve the data exchange bottleneck?
Most importantly, how do we confirm that what Professor Lin said is true?
Could this be a conspiracy, a conspiracy to make competitors waste energy on this?”
Ilya said: “Professor Lin mentioned in the live broadcast that papers verifying the relevant theory will be publicly published, and he will demonstrate the necessity of the left-right brain architecture.”
Jeffrey Hopkins corrected: “Yes, but the problem is, theory being true doesn’t mean it’s feasible in engineering, doesn’t mean current hardware technology is sufficient to support engineering practice.
Just like President Reagan’s Star Wars Program in the 1980s, does it make sense? Yes, if I intercept all the missiles, Soviet Union’s nuclear weapons can’t hit us, right? Theoretically feasible? Yes, as long as my guidance system is precise enough, we can do it.
That way, Soviet Union poses no threat to us at all, right?
The problem is in reality, after spending over a trillion US dollars, it wasn’t achieved, still a long way from true all-around interception.
The distance between theoretical feasibility and engineering feasibility”
Hopkins paused for a moment, then stretched out both hands in a forward embracing gesture: “Might be even larger than the entire universe.”
Amy added: “Exactly, this point is very important.
This requires a very efficient interaction mechanism. Without an efficient interaction mechanism, two clearly divided hardware units would likely become two mutually isolated systems, which is not only inefficient but could lead to more problems.
From a global perspective, what Professor Lin proposed is undoubtedly a forward-looking strategic direction.
Future AGI will need multiple types of computation to work synergistically; we can’t rely solely on the existing unified architecture to support more complex tasks.
Assuming our future models not only strengthen perception tasks, but also include reasoning, planning, emotional understanding, these tasks inherently have completely different computational characteristics.
If we can’t divide labor and cooperate at the hardware level, these tasks will drag down overall efficiency, making it counterproductive.”
Jeffrey Hopkins said: “I also believe this is not a temporary hot topic, but part of Crimson Technology’s long-term strategy.
We must also focus on future intelligent system architecture, considering deep integration of hardware, algorithms, and applications, to take the lead in the next generation of artificial intelligence breakthroughs.
But the question is, when do we start? How much effort to put in? With limited personnel now, how to achieve such architecture integration?
These are all hard to decide.
And in the past, China has always followed us; this time we can also follow China, wait until they achieve good results in this architecture, then we follow up.”
After listening overall, Sam Altman was moved.
He is not an artificial intelligence professional, he is an angel investor, an investor; he doesn’t understand technology, but he understands business.
He already heard what he wanted from the experts’ discussion.
That is expectation, nicely called expectation, bluntly called gimmick.
Sam Altman waved his hand: “This is not a problem. Intel’s Loihi chip is still a complete failure project to this day; besides research significance, it hasn’t played any role in the actual market.
As long as we propose cooperation with Intel, money, people, and computing cards will all be provided by Intel.
Nvidia dominates left brain chips with a market capitalization of four trillion US dollars, while Intel’s market capitalization is only 100 billion US dollars.
A forty-fold gap; if they can occupy the right brain chip ecosystem position, they can also reach Nvidia’s level.”
Other researchers immediately realized, yes, we can get something for nothing.
Andrew Kaplan reminded: “Exactly! Not just Intel, but also IBM.
IBM’s TrueNorth chip also tried to simulate the structure of brain neural networks. Intel’s Loihi showed excellent parallelism in perception tasks; their hardware all faces the same problem: when tasks become more complex, synergy and division between hardware becomes especially important.
We can completely use this as leverage to test whose sincerity is higher between Intel and IBM.
Just like Professor Lin said in the live broadcast, he submitted needs to four chip companies in China, and the winner becomes his partner.
We can also make Intel and IBM compete in the same way.”
With both sides bidding, the price will be high enough. Intel and IBM are symbols of America technology, real native America companies, which also means they can offer particularly high prices. Sam Altman pondered.
New Year’s Day 2025 is a Wednesday, so the next day is a workday, the day the stock market opens.
Early in the morning, China Mobile took the lead, a straight line straight to the sky, directly hitting the daily limit.
As a central state-owned enterprise with over 2 trillion RMB market capitalization, China Mobile absolutely qualifies as an elephant dancing, further driving the entire central state-owned sector into a frenzy.
“Ran Shen said he uses Mobile! How can Mobile not be good? Definitely a national brand, this valuation is still too low! Give a target, four trillion!”
“Ran Shen’s mobile phone number is Mobile’s, what does this mean? It means Mobile is deeply trusted by the official, means high reliability, political reliability is the greatest reliability! What reason do we have not to buy such a reliable company?”
In East Money’s stock bar, stock bar brothers are no different from forum brothers, all finding reasons for the result, swearing confidently.
Even without results, they can fantasize a result themselves, then say their holdings will 100% rise, no reason not to rise.
One drop and it’s the banker plotting against me, some AI monitoring my account, drops when I buy, rises when I sell, the entire big A teaming up against me.
The chip sector followed with a surge, led by Cambricon, also directly hitting the daily limit.
“Just the right brain chip concept is worth four trillion US dollars, okay! Left brain chip Nvidia four trillion market cap, right brain chip Cambricon not even 300 billion RMB market cap, one daily limit? At least 10 daily limits first to 2 trillion RMB.”
Similar posts are countless in Cambricon’s stock bar, with retail investors (leeks) weakly asking: “Didn’t Ran Shen say there are four in total, competing with Huawei too? How is what you say set in stone?”
“? If you don’t dare think this, why fry stocks! Stock trading is trading expectation, buy in divergence sell in consensus! Precisely because of competition, divergence creates buy points, funds will pour in!
If it was just Cambricon, already selected, would it be your turn to buy?
Buy when no one cares, sell when everyone clamors!
Institutions and hot money who got the news started bottom fishing 800 years ago, okay?
Where do newbies come from who don’t understand this!”
“No, this is also no one cares? Cambricon is about to explode, already had one round before, company with PE ratio to the sky, and you say it’s no one cares?”
“Four plus Huawei, equivalent to selecting one from five, Cambricon as the only listed company, doesn’t that mean it has the greatest hope? But I still hope to hear news of Crimson Technology taking shares in Cambricon.
Huawei won’t let you take shares, Cambricon can. From management to shareholders, I guarantee all raise hands and feet in approval for Crimson Technology to take shares.
Once it takes shares, it can steadily secure a seat via equity relationship, becoming a right brain chip manufacturer.
Here I call on Cambricon’s secretary of the board, quickly mobilize all resources you can leverage and contact! If not now, when? Even I as a retail investor am anxious!”
The stock bar is buzzing nonstop, everyone cheering for their own cognition and accounts.
Cambricon’s secretary of the board, on the first workday of the year, received 800 calls.
Retail investors calling in, shouting to be the final selected company.
Institutions, from Yanjing to Shanghai, from Yangtze River Delta to Pearl River Delta, any notable securities company or fund company wants to do research at their company.
No stop all day.
And the other three, investors directly went to the company, demanding an explanation.
Moore Threads not yet listed, these three not listed. Moore Threads signed a strategic cooperation agreement with Zhejiang Huashu, a subsidiary of Huashu Media, in 2022. Though no specific business cooperation, Huashu Media opened and hit daily limit.
Heertai, holding 1.244% shares of Moore Threads, also hit daily limit.
Anything even remotely related, stock prices flying to the sky.
“No, Mr. Zhang, such important news, why didn’t you tell us earlier!”
In Moore Threads’ meeting room, densely packed with representatives from various investment institutions.
Moore Threads, like Cambricon, is a domestic GPU company. Unlike Cambricon focused on AI chips, it does both B-end and C-end chips, meaning they also make gaming graphics cards for ordinary players.
Its founder Zhang Jianzhong was previously Nvidia Greater China’s CEO and global vice president. With “Zhong” in the name like Morris Chang, this resume and coincidence make much capital very optimistic about its development.
This is even more delightful.
Unexpected bonus.
The company they invested in can actually become Crimson Technology’s partner, not just remotely related, but right brain chip, such a core concept.
You can’t say the brain isn’t important, right?
As for being one of four competing with Huawei, that’s not important. To these investment institution executives, it’s as good as selected.
Man makes the difference, already only four competitors, hard to stand out among them?
In some sense, these executives’ thinking is no different from retail investors.
“It’s not that I didn’t say, but we just got the news not long ago, just first contact from Crimson Technology side in November last year.” Zhang Jianzhong said with a wry smile.
All present are financial backers, no one dares offend, including Shenzhen Capital Group, Xiao Ma Executive, Sequoia China, Shanghai Guosheng, CCB International, Qianhai Mother Fund, etc.
Mainly state-owned capital, some foreign and private capital.
“Then subsequently multiple rounds of contact with Crimson Technology, finally they confirmed we qualify to participate in their proposal competition. Now still in scheme design stage, need proposal and demo, then based on our designed chip, determine if selected.
The whole process at least another half year or so, not that fast.
I know everyone is excited, I was as excited as you all when I learned, this is extremely important for Moore Threads, even determines whether we are a mediocre company or can become a great one.
All of us are preparing with 200% effort.” Zhang Jianzhong said seriously.
The meeting room atmosphere was hot, as hot as today’s big A.
“Effort can’t just be tactical effort, but strategic effort!” Shenzhen Capital Group executive Chen Hongyi said with equal seriousness: “So many prominent figures in the investment circle here.
Mr. Zhang, if you told us earlier, we could have acted earlier.
I know your thinking, pre-managing expectations, afraid our expectations too high, if not selected, causes negative chain reaction.
You think left-right brain structure, right brain chip, definitely high secrecy level, with Crimson Technology and General Manager Lin’s secrecy requirements, and Huawei’s recent incident, even if not selected ultimately, outsiders won’t know the process.
Result unexpectedly, General Manager Lin said it himself in the live broadcast.
No, don’t you think, is this something that can be hidden? Is this the same thing?
Involves four companies plus Huawei, how can you possibly hide it?
Other institutions are already moving, shouldn’t we move?
Mr. Zhang, you are from technology background, we respect your professional, believe in Moore Threads’ R&D strength, but problem is, many times victory or defeat is off the chessboard.
Your confidence is good, but can’t let your confidence ruin the entire company’s future!”
Chen Hongyi’s words were light and heavy, but dissatisfaction was already evident.
Zhang Jianzhong smiled wryly: “We thought technology speaks, we already poached two engineers from Intel and IBM who worked on neuron chips similar to right brain chip, and simultaneously searching market for direction-fitting engineers.”
Chen Hongyi sighed: “Mr. Zhang, this is not just a technology matter.
Haven’t you got it yet?
Victory or defeat not on the chessboard.
Biren Technology’s backer is Tencent, no need to say Tencent and Crimson Technology’s relationship?
Bluntly, the four GPU little dragons General Manager Lin mentioned yesterday, all having tough days, losing money every year.
But once become right brain chip supplier, Crimson Technology and Apollo Technology’s bullets are endless, any order can turn each profitable.
Not to mention, this means securing a seat in future artificial intelligence ecosystem.
The interests involved already not just decided by technology.
I dare say, Biren Technology definitely told Tencent, letting Tencent behind work on it.
Also Moore Threads is Yanjing company, Biren Technology is Shanghai native company, do you think Shanghai local government hopes you selected or Biren Technology?
Will Shanghai side work in between?
Biren definitely trying, more bluntly, they 100% using such off-board tactics to become right brain chip supplier, gain Crimson Technology’s support, like TSMC supported ASML in the past, OpenAI and Nvidia cooperation.
You already behind three months, good thing three months not long, we still have time.
General Manager Lin mentioning four competing in live broadcast means Biren Technology hasn’t won yet.
Good, good.”
Zhang Jianzhong even read a post-disaster relief flavor on Chen Hongyi’s face; first time seeing such emotion on this financial executive.
Zhang Jianzhong said: “Manager Chen, I believe General Manager Lin will give us a fair competition opportunity. I think as a technology company, ultimately rely on technology.
From top to bottom, all troops fully committed, determined to win.
Using off-board tactics, trying to win via interest connections, sure marketplace like battlefield, no one blames winner for morals, I understand, but this time different.
This time competing for the key to the future, I have a feeling left-right brain chip will be humanity’s key to general AGI, and we competing to be one of the materials forging this key.
This time, I precisely hope to win by hard strength, not off-chessboard wins.”
