Author | Huang Yu Editor | Liu Baodan Last year, thanks to the success of the "Speeding" on iQiyi, the company has had a difficult time recently. On the one hand, the explosively popular TV series is on hiatus, and on the other hand, the derivative concert of the variety show "Plant Some Goodness" has been criticized for "free offline but paid online." According to Wall Street News, the second Wheat Field Music Festival, produced by iQiyi's reality show "Plant Some Goodness," was held on June 6th. Some viewers had previously received free offline tickets through official activities, while online viewers, even iQiyi members, had to pay RMB 12 for viewing, and the viewing period was valid until June 14th. This differentiated pricing model has caused dissatisfaction among many viewers, who question that iQiyi's move is "cutting corners." In response, iQiyi's customer service said, "You can buy tickets to watch the concert live according to your own needs, and the edited content of the concert will be launched on the main platform in the future." In fact, this is not the first time that long video platforms represented by iQiyi have been accused of "cutting corners." In recent years, membership grading systems, early access, and inventory restrictions have often caused user backlash, in addition to paying extra for derivative programs. The differential pricing model reflects the growth anxiety faced by long-form video platforms such as "i优腾."

In today's weather is good. Today's weather is good.

In the blink of an eye, it has been two years since ChatGPT sparked the wave of AI large models, but the significant business opportunities that many anticipated have yet to arrive.

Unlike the favorable outlook of last year, the venture of AI large models is entering a tumultuous period. Since the second half of this year, there have been frequent reports of key technical personnel from major domestic AI companies and large model teams leaving, creating instability.

First, Huang Wenhao, co-founder of Zero One Technology, left, followed by Zhou Chang, head of technology at Alibaba's Tongyi Qianwen, who transferred to ByteDance. Additionally, Liu Wei, head of technology at Tencent, and Ai expert Yan Shuicheng from Kunlun Tech also departed from their original companies.

Talent mobility is a barometer for the development of the AI industry. Behind the departure of these technical backbones are multiple challenges faced by AI large models, including a slowdown in technological iteration and subpar commercialization. Everyone is actively adjusting, searching for what they believe to be the correct path or direction.

The AI large model industry, which is currently in a state of confusion regarding its development, is experiencing a reorganization of technology, capital, and talent, with a reshuffling quietly occurring. The trend of industry consolidation will become increasingly evident in the future.

In fact, such stories are played out in every wave of technology. An industry consensus is that after intense competition, only a handful of large model companies that play important roles in the future will remain, and only those who have deeply participated in the development of unicorns can become the last lucky ones.

This is a brutal competition with unpredictable outcomes. Being involved means giving it your all.

Turmoil.

In the competition for cutting-edge technology, talent can be said to be the most important competitive advantage. In the rapidly evolving field of ai large models, talent is a crucial factor determining whether the underlying technology and products can keep pace and ultimately rank among the top tier of the industry.

Several investors told Wall Street news that in this wave of ai large models, what investors value most in evaluating investment projects is still the talent team, as this determines whether there is a continuous capability for technological iteration.

However, whether in large companies or ai ventures, the talent that gathered during the initial excitement is now making new choices, either actively or passively, due to the impact of reality.

According to confirmation from Wall Street news, Liu Wei, an outstanding scientist at Tencent and one of the heads of Tencent's Hunyuan large model technology and the AI Lab's computer vision center, has recently left Tencent. It is reported that Liu Wei has started a venture in Singapore, still targeting the field of video generation.

Recently, kunlun tech also announced that Yan Shuicheng is no longer the director of its 2050 Global Research Institute and has been appointed as an honorary consultant of kunlun tech. As an expert in computer vision and machine learning, Yan Shuicheng only joined kunlun tech in September last year, helping build the 2050 Global Research Institute from 0 to 1, conducting in-depth research around the next generation model architecture and Agent.

In this wave of talent turmoil, more people are choosing to move from ai ventures to large companies, or transitioning from one large company to another.

Currently, ByteDance, which is actively preparing a large model research institute in the second half of the year, is the biggest winner in this wave of talent movement.

After the resignation of Qin Yujia from Wall Faced ai, it was reported that he joined ByteDance’s large model research institute in the second half of 2024; in August of this year, Huang Wenhao, co-founder of Zero One Technology, joined ByteDance's model algorithm team seed, reporting to Zhu Wenjia, the head of ByteDance's large models; Zhou Chang, the technical head of Alibaba's Tongyi Qianwen large model, was also reported to have joined ByteDance in October.

It is worth mentioning that Zhou Chang's departure has also sparked a lawsuit. On November 13, it was reported that Zhou Chang violated a non-compete agreement, and Alibaba has filed a labor dispute arbitration application.

Not long ago, Yang Zhilin, the founder of Dark Side of the Moon, remarked on the phenomenon of some talents returning to large companies, stating that this is normal. "The industry has entered a new stage of development; it has evolved from having many companies at the beginning to now having fewer companies involved, and the things everyone will make next will gradually differ. I think this is an inevitable trend."

The training of large models requires significant investment, and even large companies must make choices. The emergence of the 'text-to-video' model Sora earlier this year once stirred up a global competition for ai video generation. However, OpenAI announced a delay in Sora's updates due to a shortage of computing power, resulting in it not being available to the public yet.

Clearly, without a clear application scenario and commercial returns, 'Sora-like' video generation models will not become a key focus for Tencent. Against this backdrop, Liu Wei, who aims to make a mark in the field of video generation, naturally needs to seek other opportunities.

A person in charge of investments at a large domestic company told Wall Street Insights that overseas, talent movements have also been frequent this year, mainly due to ai large model teams facing short-term technical bottlenecks and challenges with slow commercialization. In the future, many domestic ai startups may face the fate of funding chain breakage and being absorbed by large companies.

Additionally, Shen Meng, the executive director of Xiangsong Capital, pointed out to Wall Street that behind the frequent movement of talent, on one hand, is the lack of deep research and development innovation in domestic large models, which leads to smaller barriers for personnel movement between teams; on the other hand, it also reflects industry restlessness and the existence of a bubble in the number of models.

Future

Since the emergence of ai as a discipline, it has been over 60 years, during which multiple technological waves have emerged, and in the early stages of some of these waves, it was as hot as the current large language models. In the initial phase of the ai wave that surged in 2016, technology companies also tried everything to grab top ai talent.

However, the patience of capital is far from sufficient to support the research of ai scientists. When ai technology fails to bring commercial realization for a long time, both major internet companies and star ai firms begin to return to rationality and reassess the 'value' of ai talent, resulting in even faster talent movement.

History is always similar, after a period of enthusiasm, the ai large models industry will also enter a phase of contraction and clearance.

In October 2022, ChatGPT triggered the global wave of ai large models, which sparked a battle among a hundred models in China, with venture companies rising up like mushrooms after rain, and major internet firms joining in, declaring the slogan 'All in AI'.

However, after a year or two of exploration, more and more companies have deeply realized that the fortunate ones who can survive until dawn are just a minority.

Baidu founder Li Yanhong also openly pointed out that, just like many previous technological waves in history, after passing the initial excitement stage, the technological bubble of generative ai is inevitable. Then, when this technology does not meet the high expectations of the initial excitement stage, people will feel disappointed.

Li Yanhong predicts that during the stage of ai bubble bursting, those pseudo-innovations that cannot meet market demand will be eliminated. After this, only 1% of companies will stand out, continue to grow, and create great value for society. "Right now, we are just experiencing this phase; the industry is calmer and healthier than last year."

All ai large model teams are at a crossroads of making trade-offs.

The most exciting event in the domestic large model industry during the first half of the year was the price reduction wave. According to Zhang Peng, CEO of Zhizhu, this phenomenon is because everyone is unable to find differentiated value points and can only compete on price.

Zhang Peng revealed that recently he has seen many industry-leading companies developing their own large models starting to turn back, because they realize that this is not as easy as it seems; it's not just about forming a team and running an open-source model. It might be better to procure.

Additionally, in early October, there were reports in the market stating that among the six companies known as the "six little tigers of AI," including Zhizhu AI, Lingyi Wanwu, MiniMax, Baichuan Intelligent, Yue's Dark Side, and Jielue Xingchen, two companies have decided to gradually abandon pre-training models, reducing the size of their pre-training algorithm teams, and shifting their business focus to ai applications.

Yang Zhilin believes that there is still half a generation to one generation of space for pre-training, which will be released next year. Next year, the leading models will push pre-training to a more extreme stage. The most important next step is reinforcement learning; it is still Scaling but achieved through different methods.

At the same time, Yue's Dark Side also actively chose to reduce business operations, focusing on perfecting a single business product. Yang Zhilin revealed that Yue's Dark Side will determine based on the usa market which business has a higher probability of becoming substantial. Focusing on the issues with the highest potential should also be most relevant to the mission of AGI.

The ultimate goal of ai research is to achieve general artificial intelligence (AGI).

"Rome is always there, but the road taken is different." Recently, Kang Zhanhui, director of Tencent's machine learning platform, stated that while everyone is considering AGI, the next two to three years can be relatively well planned, but the routes taken by different companies may vary. For example, Tencent has chosen to follow the path of a mixture of experts model (MoE) structure.

However, regardless of the chosen route, a common challenge faced by all AI entrepreneurs is that high computing power brings high costs, and there is currently no commercial realization path that can cover these exorbitant costs in the short term.

Shen Meng, executive director of Xiangsong Capital, told Caixin that large models will soon enter a painful elimination phase, and those technologies and products that push towards core technologies are more likely to gain market recognition.

This is a once-in-a-century technological revolution, but without sufficiently mature technologies and reliable business models to support it, AI large models are likely to end up like VR and the metaverse in recent years, where the frenzy subsides and leaves a mess.

The elimination phase has now begun. Before the 'iPhone moment of AI' arrives, all companies need to demonstrate enough patience and high sensitivity to meet the brutal challenges ahead.