Chapter 345: Life Rarely Offers Several Chances To Go All In
“Lin Sheng, this is just too unbelievable.
How did you do it?”
After Lin Ran demonstrated, Tencent’s executives each took turns trying it out, and the capabilities shown were even more impressive than their most optimistic estimates.
Impressive to the point that they even suspected whether there was an entire team providing answers in the background.
Because GPT is very powerful, but its performance is more pronounced in English contexts; in Chinese contexts, GPT’s performance has its limitations.
The main issue is that the responses lack soul, as if it first translates your question into English, then internally responds in English, and translates the response into Chinese for output.
This leads to the user experience of conversing in English versus conversing in Chinese with GPT being two completely different worlds.
For Tencent’s executives, their most optimistic estimate was that the Chinese user experience would be about the same as GPT’s.
After all, in just over two months, GPT-1 and GPT-2 were open-sourced, but GPT-3 was not, so you couldn’t use others’ open-source methods to train it.
Moreover, even just GPT’s Chinese performance would be sufficient for Tencent, since GPT can’t enter China, and their artificial intelligence isn’t counting on going to America either.
Just like in the past, the Chinese Internet is one world, and the external network is another world.
But its performance feels like it has a soul.
Especially in terms of Chinese literary quality, it demonstrates extraordinary literacy.
Much more powerful than GPT-3.
Once the executives left and only Lin Ran and Pony remained, Pony could hardly conceal his inner excitement.
How could he not know what this means for Tencent.
It means that among the BAT trio, B is about to be eliminated.
Lin Ran said: “Pony, the technical route is already established, the resumes of GPT’s R&D personnel are public, their published papers are public, and the past GPT-1 and GPT-2 are open-sourced.
Actually, through these, you can see exactly how they did it.
What methods were used, what fields’ knowledge it involves, how the overall architecture is designed.
All of this is clear as day from the existing information.”
Lin Ran wore an old-fashioned gray cashmere sweater, a faint smile on his face, as if this thing more powerful than GPT-3, Crimson, was insignificant.
Pony was utterly convinced and admiring, saying it’s easy but doing it is another matter entirely.
He dared to guarantee that if anyone else took over, even if they poached OpenAI’s chief scientist, they couldn’t achieve this effect in just over two months.
As for calling it Crimson or Alpha, that doesn’t matter.
If you can create it, even if you call it Deep Blue, I’d only applaud and say it’s a good name.
As for whether Crimson will trigger external network public opinion, or remind people of the red tide making a comeback, that doesn’t matter.
Pony never believed that China’s artificial intelligence could dominate in America-led European and American countries anyway.
“Lin Sheng, so shall I plan a press conference in March?
Tencent will mobilize the entire company to fully promote this product.” Pony looked excited.
The timing is crucial.
At this current time, Tencent’s market capitalization on the Hong Kong Stock Market has been plummeting.
From the peak in February 2021 when it was over 700 Hong Kong Dollars per share, all the way down to last November when it hit a low of less than 200 Hong Kong Dollars per share.
That’s not even a halving; it’s been chopped down to the ankles.
Chopped to the point that Tencent was even forced to spend its own money on stock buybacks.
There are various reasons behind this, but more fundamentally, Tencent has lost its imagination.
What does the capital market rely on? Imagination.
Whether Nvidia is worth four trillion US dollars, whether the AGI path can truly lead to artificial general intelligence, and whether Nvidia is truly indispensable in that— these are all unknowns.
But the key is, if the capital market believes in you, your stock price can soar.
When you’re at one trillion, someone shouts two trillion; when you reach four trillion, the market starts shouting eight trillion, even ten trillion US dollars.
Tencent’s fundamental problem is that it has no more imagination, while ByteDance is seizing people’s time, energy, and attention.
Tencent’s throne as the top in traffic has been taken away.
WeChat’s position certainly won’t be shaken, but WeChat’s ecosystem lacks the imagination of the short video ecosystem.
The emergence of Crimson means new imagination has appeared; Tencent possesses the only domestic artificial intelligence that can compete with OpenAI, surpassing Google and META.
Pony knew very well that once this deep-sea bomb is dropped, Tencent’s decline will be swept away.
Lin Ran nodded: “Pony, you decide on this.
Now we need to clarify the subsequent cooperation model for Crimson.
And the potential risks.”
Pony understood the other’s concerns immediately after hearing this.
Having stood firm in China for so many years without falling, even Jack Ma has stumbled and said things that offended people.
Pony has also spoken publicly many times in the past, yet he never had such mishaps.
Even if there were occasional oversteps, Pony could keep the risks under control, showing his political wisdom.
“You mean pressure from the White House?” Pony asked.
Lin Ran said: “Exactly, Tencent’s business spans the globe, and AGI is a lifesaver for America.
For a Chinese company to launch an artificial intelligence more powerful than GPT in just two months is unacceptable to the White House.
They will use every means to suppress Tencent.
They can’t influence me personally, nor Apollo Technology, so Tencent is bound to be the target bearing this pressure.
You need to think this risk through clearly.
Tencent is listed on the Hong Kong Stock Market, out of the White House’s reach, but Tencent has minority businesses and many investments in America, all of which could be affected.”
Pony said firmly: “In front of a large model, those are all expendable.
Besides, America can’t just confiscate them outright; even if for security reasons they disallow our investments, they’ll give time to exit.”
Pony saw it clearly; at this time point, these are minor details.
Lin Ran continued: “There’s also security, meaning personal safety.”
Pony asked: “You mean my children? They’re all domestic; I’ll instruct them later not to run around unnecessarily, and if they need to go abroad, to entrust others to handle it.”
Lin Ran nodded: “Not just you, but other Tencent executives too; this needs to be emphasized.
Though it’s an extreme case, America might resort to desperate measures.
Before, the Apollo Moon Landing got my mentor imprisoned along with it, and our connection was as weak as could be.
So risks in this area need to be prevented in advance.”
Pony sighed deeply, “What a terrible era; our generation grew up watching American movies, accepting America as the lighthouse, using American software.
Watching America degenerate to this point, unable to win through normal competition, resorting to all sorts of unconventional means.
Lin Sheng, I understand; I’ll definitely take this seriously.
After all, it concerns personal safety.”
Lin Ran said: “Pony, no need to worry too much.
Personally, I think America is definitely off-limits in the short term; you can only entrust professional managers to handle it.
But other European and American countries are fine; after fully communicating with international lawyers to ensure no handles to grab, you can still go.
After all, America still needs some reason.”
Pony smiled bitterly: “Do they really need a reason?
Like you said, your Indian mentor—I saw the news too; now he’s neither seen alive nor dead, and when asked, he’s just ‘under federal investigation.’
As for where exactly the investigation is, why lawyers can’t meet him, has America ever given a straight answer?
And that’s for an Indian; we’re Chinese people.”
Hearing this, Lin Ran also sighed deeply.
Because Pony was absolutely right.
“Lin Sheng, regarding the specific cooperation method, my idea is, would you consider Tencent acquiring Crimson outright for cash?
I’m willing to offer fifty billion US dollars to fully acquire your shares.”
Fifty billion is undoubtedly a massive move among massive moves; Tencent’s current cash reserves are roughly this amount.
But not exaggerated.
OpenAI’s valuation is fifty billion, and for Crimson, which matches it, this valuation isn’t high.
Especially since it’s full ownership.
Lin Ran asked: “Pony, are you sure that without me later, Crimson can keep up with GPT’s progress?
An outright acquisition would be a one-and-done deal.”
Pony said: “Of course I have no confidence, so after full acquisition, the chief technology officer will still be you; I’ll offer you an extraordinary annual salary.”
Lin Ran waved his hand: “Full acquisition is out; instead, Crimson and Tencent sign a deep exclusive authorization agreement, with the backend model being Crimson, and Tencent acting as the frontend entrance.
I plan to raise fifty billion US dollars in cash from the market, giving Tencent a ten billion investment quota; we need full cooperation with Huawei.
Based on our needs, to develop AI computing cards.
This portion of money is mainly for breakthroughs at the hardware level, to build our moat.
Huawei’s current Ascend is clearly insufficient; algorithmic advantages can compensate for computing power disadvantages, allowing us to match OpenAI within two years.
But over a longer time frame, they’ll improve algorithms too; our engineers will be poached, our algorithms can’t stay secret forever.
Ultimately, it comes down to internal energy.
We need to burn a large sum of money, cooperate with Huawei, push them to develop a dedicated Crimson computing cluster tailored to our needs together.”
Pony understands politics, but Lin Ran understands it even better.
Right now, you can’t give all the investment shares to just Tencent, only you two.
After pondering for a moment, Pony said: “Okay, Lin Sheng, if it’s fifty billion US dollars for Tencent to fork over all at once, I might have to grit my teeth and pay in quarterly installments to your account, but if it’s just ten billion, Tencent’s money can hit the account anytime.”
Money is just numbers; cash sitting in accounts is meant to be used.
But fifty billion US dollars at once would pressure Tencent; having that much cash doesn’t mean you have to spend it all in one go.
At this time point for Tencent, cash and cash equivalents on the books add up to about exactly fifty billion.
Pony deeply agreed with Lin Ran’s plan to invest in partnering with Huawei for computing cards: “In the last two months, because of LLMs, not just Nvidia’s computing cards—even Huawei’s Ascend is in short supply.
Apollo Technology locked in most of Ascend’s production capacity in advance, causing Huawei’s Ascend shipments to be severely insufficient; compared to our order price back then, prices have risen over 50% now.
All major companies are building AI computing power infrastructure.
We really need to do this.”
Pony even thought Apollo Technology was too lucky, locking in Huawei’s production capacity early.
Lin Ran continued: “I only need Alpha’s control; I need to ensure the entire technical route and future developments follow my vision; I need voting rights.
So for voting rights design, I’ll adopt an AB share structure.
Crimson is exclusively authorized to Tencent, operated by Tencent.”
Before this, Alpha Technology’s share allocation was Lin Ran at 70%, Tencent at 30%; Tencent’s early provision of computing power and personnel counted as costs.
Now needing to raise another 500 billion from the market, Lin Ran’s shares would be diluted; at this point, only through AB shares—same shares different rights—can control be ensured.
Pony didn’t quibble, “I agree in principle, but I need to go back and hold a board meeting first.
With Crimson, convincing the directors won’t be hard.
As for other state-owned capital, I believe it’s no problem for them; they lack promising projects, not money.
Besides, you’re cooperating with Huawei, wanting RMB; for state-owned capital, RMB doesn’t even make them blink.”
After witnessing Crimson firsthand, the pressure from GPT vanished instantly without a trace.
Pony felt fortunate inwardly; his luck was too good—from the PC era to the mobile Internet era, and now to the artificial intelligence era, Tencent always gets the ticket each time.
After seeing Crimson’s miracle in just two months, Pony didn’t believe any domestic manufacturer could catch up to them in the GPT field.
“Lin Sheng, one more thing: on specific pricing, how should we settle between us?” Pony continued, “I think we two talking directly is more straightforward than our subordinates.
Tencent using Crimson’s services can’t keep mobilizing at no cost, and Crimson using Tencent Cloud’s infrastructure can’t be unconditional forever.
Since we’re two companies, we have to separate accounts.”
Lin Ran said: “Pony, that’s too detailed; it’s a business matter. Once the company’s framework is set up, we’ll clarify these properly.
So initially, Tencent still needs to send a professional manager to work at Crimson.
His first task upon arrival is to rename Alpha Technology to Crimson Technology.”
Pony was truly curious: “Lin Sheng, I feel you’re particularly attached to the name Crimson; why is that? Is there something special?
If it’s inconvenient, no need to say.”
Lin Ran thought to himself that the real reason was inconvenient, but on the surface, what inconvenience: “Pony, it’s like this: IBM is the Blue Giant, and their first chess program was called Deep Blue, right?
So I thought, calling ours Crimson—DeepRed—fits China’s characteristics very well.”
Pony laughed: “Compared to Alpha, Crimson indeed has much stronger Chinese characteristics.”
He then brought up: “Lin Sheng, as a professional, you tell me: can LLMs really lead to AGI?”
Lin Ran thought for a moment, his gaze passing through Pony to outside the window; the falling rain and snow reminded him of the season, and he said: “I believe that before there’s a revolutionary breakthrough in hardware, there’s an insurmountable high wall between LLMs and the artificial general intelligence promoted by OpenAI.
LLMs are like a super librarian; they remember and recombine patterns from countless books to quickly assemble answers for you.
It doesn’t truly understand; it just works on statistical probability.
For example, if you ask it a history question, it might answer eloquently, but if the data is wrong, it’s prone to nonsense and hallucinations.
AGI: This is the ‘all-purpose AI’ from sci-fi, able to autonomously learn new skills like humans, reason about unknown problems, even invent things.
It’s not limited to text; it can handle all real-world challenges, like driving, doing experiments, or emotional interactions.
My personal view of LLMs is that their limit is a super useful assistant, a search engine that can help people improve efficiency, but absolutely can’t replace humans.
I even think the shackles limiting from LLMs to AGI are not just algorithmic, but also hardware-level; with the current architecture, achieving AGI short-term is unrealistic.
Although in terms of transistor count, we can reach the same order of magnitude as human brain neurons, transistors only have two states: 0 and 1, but neurons can connect to thousands or tens of thousands of synapses, with total synapses on the order of 10 to the 14th to 15th power.
I think this is the more essential reason why silicon-based chip artificial intelligence can’t emerge true wisdom patterns.”
Once the bosses decided, it was time for the underlings to work.
Zhao Songxia’s Enterprise WeChat kept dinging nonstop.
First, in the company group, the team leader sent a message @everyone:
“The campaign has ended successfully; for those who can continue at Alpha Technology, private message me your resume. Not everyone can stay; at most about 300 retention spots. I’m not even sure if I can stay.”
Then the small groups, all sorts of small groups, everyone discussing this.
“OpenAI only has about a hundred people; a company like this really doesn’t need so many. These two months, work wasn’t even that saturated.”
“Regret it; regret not leaving a profound impression on General Manager Lin. Didn’t even attend a single work report in these two months.”
“General Manager Lin is rarely in the office; his visits can be counted on two hands. Even if you want to report work, you have to find him first.”
“Yeah, and we do data processing; what work to report? Unless there’s some innovative method, but you have to report to the small leader first, and they’ll just steal your idea.
After all, with just your idea, General Manager Lin can reverse-engineer it; no chance for you to hold out for better.”
“Any backdoor ways to stay? Life rarely offers chances to go all-in!”
“This counts as going all-in? This is purely boarding General Manager Lin’s big ship. Ten years ago, the WeChat team lived like immortals—everyone’s seen it. Now an even more awesome big ship than the WeChat team has appeared; this isn’t going all-in, it’s doing everything to get on board.”
“Yeah, back then the WeChat team competed on internal energy; any product could be made. But now it’s different; can anyone match General Manager Lin’s mind?”
Zhao Songxia looked at the small group and found it amusing; if anyone had a way, would they tell others?
Everyone is a direct competitor.
Two thousand people, at most three hundred can stay.
These two thousand people were already the cream of the crop, selected through fierce competition to enter Tencent as AI algorithm engineers; now it’s elite of the elite.
No one can guarantee they’ll stay.
Zhao Songxia knew that at this time, he needed a backdoor move. Life rarely offers chances to go all-in—that’s right, but not boarding the ship by going all-in, rather going all-in for a chance to board.
“Hey, Xu Xian, you there?” After going home, Zhao Songxia called Xu Xian.
This was his only connection, one that could reach Lin Ran directly.
There was once a so-called six degrees of separation theory, meaning through six people, you can connect to anyone in the world.
Zhao Songxia was glad his Yenching University mathematics department undergraduate experience could come into play again.
Back then, entering Tencent as an algorithm engineer relied on that resume.
In 2018 when job hunting, the environment wasn’t as harsh; he’d just learned some artificial intelligence and algorithm knowledge, and rode on Yenching University mathematics department’s gold signboard to enter Tencent.
Now he had to rely on that experience again.
“Songxia-kun, what’s up? Brother Ran giving you trouble? Want me to put in a word?” Xu Xian answered without politeness.
Xu Xian was Xu Xian’s nickname; Songxia-kun was Zhao Songxia’s nickname, implying he was Japanese.
After all, Japan likes naming with place names; an old Chinese Internet joke is that wherever Japanese people work, they name their kids after it.
Like Yamamoto, Tanaka, Kawaguchi, Murakami, etc.
Zhao Songxia’s name sounded very Japanese, so they liked adding -kun.
Xu Xian and Zhao Songxia were undergrad roommates; later Xu Xian pursued a PhD, Zhao Songxia went to Tencent, but they still chatted occasionally, good relationship.
After Xu Xian posted a photo eating with Lin Ran, Zhao Songxia learned his college roommate had such a connection, actually high school classmates with Ran Shen.
At the time, he also followed the trend in that Moments post:
“Ran Shen, take me flying!”
Later, when Zhao Songxia was transferred to Shanghai on business and worked under Lin Ran, he stealthily took a photo of Lin Ran and sent it to Xu Xian, sharing the experience of working near Lin Ran.
So that’s why Xu Xian asked that.
“Of course not; between General Manager Lin and me, there are at least eight hundred levels difference; how could he trouble me.
It’s like this: the campaign ended, and I want to stay working by General Manager Lin’s side.
See if you can help me out, pass a word to General Manager Lin.” Having rolled in the workplace for so many years, Zhao Songxia said it without any embarrassment.
Connections must be used when needed.
Staying through connections is also a skill.
“Huh? Isn’t staying at Tencent way more promising than by Brother Ran’s side? No matter how awesome Brother Ran is, it’s just an aerospace company doing aerospace-related business.
Salary ceiling is very low; you got charmed by Brother Ran’s personal charisma?
Personal charisma can’t be eaten.” Xu Xian teased.
Zhao Songxia complained: “No, this company is newly established by General Manager Lin and Boss Ma, mainly for artificial intelligence R&D, targeting OpenAI; it’s extremely powerful, and existing achievements are no worse than GPT.”
