Alas, this is not the first time Ali has done such a thing. Before Lin Junyang, Zhou Chang was the soul and technical leader of the Qwen team. He left his job in July August 2024 to join ByteDance, which caused quite a stir in the circle at that time. After adding ByteDance, Zhou Chang was responsible for the multi-modal interactive research and development of the Doubao big model. He is one of the core members of Seed, a fundamental research team for the Byte Big Model. The rise of Lin Junyang is directly related to the departure of Zhou Chang. After taking over the responsibility of Qwen, he led the team to create milestone works such as Qwen 2.5 and Qwen 3 series, which are the most widely used open source models in the world. Only the Qwen series truly achieves full open source in multiple languages, sizes, and modalities. Although not the strongest open source model, it is fully compatible with small scenarios, performance limited devices, and languages. Based on this, I think Apple is most likely to cooperate with Alibaba, not just in China but in many small language countries around the world. As someone who still holds Alibaba stocks, I am really hesitating now because the outdated commercialization process has forced the original core technologies away. It is difficult to say whether I can continue to lead in the open source field in the future. I am thinking from several perspectives. 1. After Zhou Chang left, Alibaba's talent reserve capability was strengthened by Lin Junyang taking over, and Qwen became even stronger. There will be stronger people to replace Lin Junyang's work. Whether this person can be competent or not is still unknown, but it is highly likely that they will not fall behind. It is almost certain that they will fall behind briefly, and this is already a fact. 2. The departure of the person in charge is indeed due to objective factors. The Qwen series has indeed fallen behind after the release of new domestic models, not as strong as GLM 5 and minimax 2.5. The only strength is multimodality, but Doubao also has multimodality and is ahead of Alibaba in terms of capability. Previously, the Qwen series may have been ahead of domestic models by 3-6 months, but now it is not only being tied but also surpassed, which is unacceptable. 3. From the perspective of Alibaba's commercialization, open source has allowed Alibaba to earn enough reputation, but it has not made enough efforts in actual monetization. ByteDance's volcano engine operates on a closed source model, and in the case of outdated capabilities, its ability to sell tokens is not much worse than Alibaba's. So in the future, Alibaba may switch to closed source models, but this premise is that your model must be ahead, so that everyone has to use it on Alibaba Cloud. 4. From the perspective of model development, the current development of large-scale models has entered a bottleneck period. Starting from this year, we will shift our focus from innovative exploration to practical implementation. The aggressive promotion of Qwen can also be seen as the current mainline, but it is possible that Qwen will completely fall behind, as the same thing has happened on Kimi. In today's AI iteration speed measured in weeks, the stagnation of underlying capabilities often means being rapidly abandoned by the market. 5. The use of DAU to assess the KPI shift of the basic model team has led to the collective departure of teams that have made significant contributions, and Alibaba has gradually become Baidu based. The reason why Baidu's model has been so garbage is that KPI oriented has achieved the worst results. From the current information, the disadvantages outweigh the advantages, affecting the confidence of holding Alibaba. The only thing to rely on is the integration of Alibaba and Apple AI, but it is still unknown.