As previously mentioned, entering the elite tier of AI research and development in China necessitates the presence of at least four key elements:

1. Possessing foundational R&D capabilities for large models.

2. The ability to continually adapt to emerging trends.

3. Support from C-end applications with hundreds of millions of daily active users.

4. Maintaining an objective and altruistic grand ecosystem.

However, this is merely the threshold for entry into this elite tier.

Entry into this tier does not guarantee leadership in the industry. Just as anyone can practice Buddhism, not everyone can attain enlightenment and become a Buddha.

According to traditional concepts, a mature market should adhere to a 7:2:1 structure, where the leader commands 70% market share and abundant resources, while the second and third players hold 20% and 10% respectively, with the remainder distributed among others. However, the AI race is protracted, and the large model race is still in its nascent stages. Maturity is at least a decade or two away, and the 7:2:1 structure will not emerge for at least 10 years.

So, readers may wonder, who stands a greater chance of winning?

Predicting an outright winner today would be pure speculation. Fortunately, while we cannot foretell the future, we can analyze it. What we need to examine is, beyond the shared 'four elements', who possesses a distinctive personality, and is this personality significant enough to become a game-changer?

When discussing personality, ByteDance must be included, as its personality value is quite high.

Why do I make this assessment? 'ByteDance's AI personality' deserves a comprehensive review, but I will focus on one aspect here – its approach to spending money.

These top players have all announced investments of no less than 100 billion yuan in AI. However, ByteDance still displays a unique trait – it spends money without stringent assessments.

Not long ago, ByteDance's Seed (large model department) held a meeting for all employees. A crucial design decision at this meeting was to completely separate the team responsible for model development and application from the team dedicated to AI basic research and exploration.

From then on, two leaders took charge of these two distinct teams. To emphasize their equal status, ByteDance's official announcement used the phrase 'the two leaders hosted the meeting together for the first time'.

Chinese culture places great importance on legitimacy, so the meaning of 'hosting together' is clear – it signifies equal status, no subordinate relationship, and cooperation as needed.

Crucially, for the team focusing on AI basic research, besides emphasizing 'encouraging team members to explore long-term, uncertain, and bold AI research topics,' another condition was added: 'abolishing quarterly OKRs and half-year assessments for this team to ensure a long-term and stable research environment.' This is a first in China's AI field.

Firstly, what are 'long-term, uncertain, and bold' research topics? In essence, they require long-term investment with uncertain outcomes and significant expenditures.

More importantly, many powerful institutions emphasize that their specific institutes or R&D centers focus on 'underlying basic research,' but none have explicitly stated 'abolishing assessments.' (Of course, assessments are necessary in the long run, but their principles should differ.)

In my opinion, this is a significant symbol of the thriving AI landscape in China.

When thinking of research institutions without assessments, one might immediately recall the renowned Bell Labs. This laboratory is not only the birthplace of almost all digital technologies we use today, such as transistors, lasers, solar cells, light-emitting diodes, digital switches, communication satellites, cellular mobile communication equipment, etc. However, what is even more renowned than these inventions is Bell Labs' signature 'de-utilitarian management'.

Many may not know that Bell Labs was also inspired by another institution. As early as 11 years before Bell Labs was established in 1925, in Europe, where old money flowed, someone had created a similar institution – NatLab. This is an internal R&D laboratory established by Philips in 1914, focusing on basic research and technological innovation. It is an even earlier laboratory that emphasizes 'free and independent research without predetermined goals' and is also known as an 'innovation haven.' One of its most famous achievements is creating the technological system of ASML, today's lithography giant.

However, please do not misinterpret my words as suggesting that any company with ample funds and an institution where researchers can independently decide on research projects, progress, directions, and funding without assessments will automatically incubate cross-era technologies and become a deity.

If that were the case, it would be akin to saying that in martial arts novels, Shaolin Temple should abolish the positions of abbot, head of the Dharma Hall, head of the Arhat Hall, etc. Everyone wouldn't need to do morning and evening prayers or receive supervision from masters and apprentices. As long as there is food to eat and martial arts to practice, and the 72 extraordinary skills in the Sutra Pavilion are open-sourced, they can naturally become the martial arts leader and the number one sect in the world.

Please revisit what I said earlier – I stated that what ByteDance has done is an 'important symbol of the prosperity of AI in China,' not an 'inevitable guarantee of the success of AI in China'.

If unlimited investment and no assessment were guarantees of success, what would be the point of competition among AI enterprises? Everyone could simply compare their bank balances.

So, let's delve deeper into the nuances.

As mentioned earlier, money is not omnipotent, but it is absolutely necessary for such institutions. Crucially, this funding cannot be cut off, and it must endure for three to five decades. ByteDance, please take note.

For example, the reason Bell Labs has endless funding is quite amusing – at that time, to avoid violating US antitrust policy (profits cannot exceed 12% of costs), Bell Labs' parent company, AT&T (American Telephone & Telegraph Company), had to artificially 'reasonably increase costs,' ensuring the laboratory received an initial funding of up to $12 million. In the subsequent decades, as long as AT&T's users paid a bill, a portion of that money went to Bell Labs...

In fact, it's not that no Chinese entrepreneurs have attempted this before. For instance, Chen Tianqiao's Shanda Research Institute proposed a similar idea and worked diligently on it for a period. However, due to the gaming business's unstable cash flow compared to the telecommunications business, this institution couldn't sustain itself for less than two years.

Therefore, let's continue to believe in ByteDance's financial prowess and prepare for it to invest for three to five decades.

Next, we must discuss the rationality and necessity of establishing such a system.

Granting high-level talents a high degree of autonomy and freedom is not a byproduct of this system but the very purpose of its design.

If we examine the establishment periods of those two renowned institutions, we'll notice a common phenomenon: they were both in a stage of new technology explosion but with unfixed development directions.

For instance, when Bell Labs rose to prominence, it was during a period of cross-disciplinary explosion in physics, mathematics, and engineering. The laboratory seized the commanding heights of disciplinary integration through the combination of 'mathematicians + experimental scientists + engineers' (such as the collaboration between Shannon and the transistor team).

For example, at that time, many institutions and schemes were competing in transistor technology, but in the end, it was Bell Labs' Shockley team's technical route that prevailed. (Shockley is an interesting figure known for being petty. I'll discuss him separately later if I have the time.)

This is very similar to the current environment in which Chinese enterprises are competing in AI – everyone knows roughly where the direction is, but the competition routes are not yet fixed. However, once a company's technical route becomes the mainstream, it will reap significant benefits.

Therefore, in today's context, only 'freedom' can cope with the current state of the racetrack, which is non-solidified, multi-path, and highly competitive; and only freedom can maximize the creativity of scientists.

In such a top-tier institution where talented individuals converge, 'freedom' is a meticulously designed system, not a loophole that 'grants' researchers 'freedom'.

Secondly, I want to emphasize that the 'basic research' here must be true basic research.

I say this because China's achievements in AI application research and development are world-renowned, but its contributions to basic research and fundamental theories are relatively few – which is not embarrassing. When these previous basic research and development endeavors were undertaken, China's digital industry was still in a difficult growth stage, making it challenging to allocate sufficient resources to basic research.

However, I must say that whoever designed this architecture within the ByteDance system deserves great respect. Its brilliance lies in recognizing that the current AI competition is not just between companies like ByteDance, Tencent, Baidu, and Alibaba. Rather, it is a competition between China and the United States to determine who will become the long-term source of AI innovation, a struggle for national destiny.

If our AI competition were purely internal, it would be akin to 'meat rotting in the pot,' sufficient to compete in application directions. However, precisely because it is a struggle for national destiny, we must rebuild a solid foundation.

This is because, from the current perspective, the rules, weapons, methods, strategies, etc., of this great battle are primarily not formulated or invented by us, putting us in a relatively passive position.

And if we want to win in the long run, we must be able to influence the overall situation of this trans-Pacific competition from commanding heights. We must compensate for our deficiencies in basics. With this mindset, we cannot expect Seed to produce results in AI basic research within three to five months. Because the cycle of basic research is relatively long.

But I do believe that allocating a separate team to engage in basic research is a prudent move for Seed. Besides showing determination, once fruitful results are achieved in underlying innovations, it will represent leapfrogging progress.

So, I must reiterate that if we are to undertake this endeavor, we must pursue true basic research.

What is 'true'? It's hard to define, but I can provide an example – it should have a considerable distance from practical applications.

Take lithography machines as an example. Everyone knows ASML. But few are aware that the underlying technology and early products of ASML's lithography machines were actually born in Philips' NatLab.

Take note! – Since there was no need to consider practical implementation, the lithography machines designed by NatLab were very advanced. Moreover, they weren't just theoretical; prototypes were actually built.

But for the next 10 years, this scheme and prototype remained in the laboratory.

Why? There are various reasons, but a crucial one is that they were too far removed from practical applications and too difficult to implement.

How advanced were they? NatLab provided the drawings and documents to Philips' production department, which built the machines, but they simply didn't work.

When the scientists inspected them, it was indeed the case. The people who built them didn't understand the principles at all; they were just 'blindly made'.

In the end, NatLab was forced to send a working group composed of scientists to guide the factory. These two sides often clashed and looked down on each other, going through countless frictions before the machines were finally built.

Even the machines built this way didn't meet the scientists' preset expectations. It wasn't until these scientists were packaged and sent to ASML, and then iterated twice, that they produced the 'relatively ideal' lithography machines in the eyes of scientists.

You might be frowning and asking, what should ByteDance do if its basic research team cooperates with the application development team as difficultly as this?

But let's first consider the outcome – ASML's machines have since dominated the market, completely outperforming all competitors with a generational lead, ending the situation where more than ten lithography machine enterprises were competing in a narrow racetrack worldwide at that time. From then on, the world only knew ASML.

The reason behind this lies in one word – 'distance.'

Listening to user feedback is the business of the product department, not the business of scientists. The profession of scientists is akin to cultivating immortals; they need to stay away from secular and lively things to listen to the voice deep within and comprehend the inspiration from the ocean of physics or the starry sky of mathematics.

Had the scientists at NatLab not aspired for the 'ultimate' lithography machine, but instead been overly concerned with the specific industrial chain conditions of the time and a realistic assessment of current front-line production, they might have improved the lithography machine, but they would never have 'reinvented it' – thereby sparking fundamental changes and developing products that, once realized, would surpass all others.

It is inherently challenging to grasp the management scale within such an institution; yet, without attempting, we will never possess such an institution or such a fundamental research team.

ByteDance has taken this leap. Though it's merely the abolition of half-year assessments, I hope it can still manage effectively and produce results, progressing from half a year to one year, until assessments are forever absent.

Now, we arrive at the final and most paradoxical point: these laboratories that are 'remarkably free and do not demand immediate results' coincidentally yield numerous achievements.

Bell Labs, for instance, has nurtured nearly 10 Nobel laureates, 7 Turing Award winners, and various other honorees, amassing over 30,000 patents. NatLab too has achieved countless milestones, including defining and commercializing light bulbs, LEDs, CDs, X-ray machines, and more, which are ubiquitous. They have also set standards for various consumer electronics and, notably, lithography machines – the cornerstone for producing the information age's core components: chips.

How does a lax management system align with an exceptionally high achievement output rate?

On a micro-level, I believe there may be two reasons:

First, it capitalizes on the genius of its talent – it's said that Bell Labs accepts only 0.3% of applicants. Hence, the final intake comprises elites. 'Geniuses are arrogant,' and the natural rivalry among talents serves as a potent internal motivator.

Concurrently, 'managers' also recognize the contributions of each researcher. For example, the 1947 Nobel Prize-winning paper on transistors was signed by only three individuals, but the engineering team of over 200 people behind it was acknowledged by the internal honor system. This is why Bell Labs has always been a collective effort.

Second, the power of example is immense. With patents emerging almost daily, Nobel Prizes being won consecutively, and generations of scientists creating world-changing products, this place has become a haven for scientific talents striving to establish their careers and impact the world. Working in such an institution has become the aspiration of top talents – scientists are not indifferent to fame and fortune but are drawn to prestigious accolades like the Nobel Prize and Turing Award. If an institution dares to claim that the probability of winning a Nobel Prize increases by 50% if you conduct research there, would it fear top talents not flocking there like fish to water? Thus, a talent loop is formed.

On a macro-level, I believe these two predecessors also possess strategic-level system designs and operating principles worth studying and learning by ByteDance – an institution that averages a Nobel Prize every three years is certainly not determined by one or two factors alone. True breakthrough innovations necessitate long-term, purposeful investments coupled with a tolerance for failure, long-term focused and high-pressure R&D, and a 'container' where diverse paradigms of disciplinary intersections can coexist to continuously generate inspiration. Of course, there's also the research freedom from short-term commercial pressures and a steady cash flow from the parent company.

Achieving all this is arduous. Yet, ByteDance is already the most qualified company.

We anticipate ByteDance's success and must tolerate its trials and errors... But we should not place all the compensatory psychology and pressure for the lack of AI basic R&D in recent years or even decades on ByteDance – it originally didn't need to impose such pressure on itself, but it did, demonstrating its pursuit. Thus, we can also observe traits and characteristics that may differentiate it from other giants, hinting at even greater potential and a higher likelihood of future success.

Therefore, I sincerely hope these characteristics can guide ByteDance down a unique path to success, benefiting both the company and China's AI industry.