DeepSeek engineer talks open source in Wuzhen: bringing AGI sooner, the opponent is time, not competitors

“We want AGI (artificial general intelligence) to arrive sooner.”

On November 3, during the AI dialogue session at the 2026 World Internet Conference Wuzhen Summit, DeepSeek research engineer Gao Xi explained DeepSeek's open-source logic. Regarding open-sourcing research results, he said it is to make AGI arrive sooner.

Against the backdrop of accelerating AI development, Gao Xi believes the opponent is time, not competitors. “In the AI race, the real opponent is not other large-model companies, but the passage of time.” He believes that no matter which company develops AGI, the technological changes it brings (such as medical breakthroughs) will benefit all of humanity. From this perspective, “when AGI arrives, you won't care which model company brought it about.”

Based on this original intention, open source has become the better choice. Gao Xi said that during R&D, after discovering new methods such as lower-cost training approaches and better model architectures, the most natural thing to do is to disclose them immediately for verification and application. If others solve a problem, then quickly follow up and apply it; if we solve a problem, then share it to advance the overall process. On October 20, DeepSeek announced that it would fully open-source its internal inference framework DeepInfer.

Liu Da Yiheng, head of the Qwen large language model project at Alibaba's ATH business group, said that open source has two major significances. First, it accelerates AI innovation and changes the innovation paradigm. Liu Da Yiheng said that open source has changed the paradigm of AI innovation, so that core technologies are no longer held only by a few companies, but developers and the ecosystem worldwide participate in innovation together. At the same time, open source is not only about opening the right to use models, but also about opening the right for everyone in the world to participate in creating AI. In addition, open source can empower academia while also giving back to the ecosystem and driving model development.

Second, open source realizes decentralization and distributed innovation. “Leading companies may overlook issues such as minority languages, special vertical applications, and adaptation to special hardware/chips. Open source allows more participants to contribute technology and form a more complete closed loop of the technology ecosystem.”

Regarding AI inclusiveness, Liu Da Yiheng said the existence of open-source models prevents price monopolies: the existence of open-source models pushes the cost-effectiveness of models to a relatively high range, avoids the price monopolies that may occur with closed-source models, and lowers the threshold for public use. In addition, in terms of data security and autonomy, for scenarios where data cannot be transmitted externally, open source allows users to deploy locally, ensuring data security.

Open-source models also carry risks, mainly reflected in irrevocability and reliance on a lack of external protection. For the governance of open-source models, the current common practice is tiered management based on model capabilities. However, this governance approach based on judgments of current capabilities has significant flaws, including underestimating model potential and the uncertainty of technological evolution. In this regard, Liu Da Yiheng said that given that restricting model openness alone is difficult to fundamentally solve the problem, it is recommended to seek solutions beyond the model, including improving laws and regulations, formulating industry standards, and building more infrastructure to mitigate security risks.