DeepSeek engineer defends open-source push at Wuzhen: 'the real competitor is time'

A DeepSeek engineer used a panel at the World Internet Conference in Wuzhen this week to defend the company's open-source strategy, conceding that releasing model weights can make safety problems worse while arguing that the race to advanced AI is one against the clock rather than against rivals. The South China Morning Post reported the remarks on November 4; the panel itself took place on Tuesday.

Gao Xi, an engineer at the company, said open development drives shared innovation and technical breakthroughs. He framed progress in the field as a matter of time. "If AGI is going to arrive in 10 years, then I hope we can bring this forward to one year. If it is going to arrive in one year, then I hope it can arrive tomorrow," he said, in rare public remarks from a DeepSeek representative.

He put it plainly: For AI model companies, the real competitor in the AI race is not other companies, but time. Technical barriers raised by competitors, he added, can only slow competition temporarily — they cannot stop AGI from arriving.

The comments land in the middle of a widening fight over open weights. OpenAI and Anthropic, the leading US labs, keep their strongest systems closed, while open-weight releases let anyone with enough hardware download and modify a model without vendor oversight. Beijing has backed open-source ecosystems as a way to work around US technology containment; security experts counter that published weights lower the bar for attackers seeking cyber or biological capabilities.

Gao did not dodge that trade-off. Open-source releases "may amplify the impact of safety issues," he said, adding that DeepSeek is strengthening safeguards including classification tiers for its models, without giving specifics.

Liu Dayiheng, who leads Alibaba's open-weight Qwen models, sharpened the point from a different angle: the "irreversible" nature of open weights creates extra safety exposure. He still rejected the closed model favoured in the West, arguing that a full governance regime and industry standards, not technological gatekeeping, are the answer. "Without open source, the core technologies of large language models would only be in the hands of a few companies," he said.

Policy is moving on two tracks at once. Authorities proposed new export controls last month aimed at frontier AI models with advanced cyber-offensive capabilities, even as senior officials kept voicing support for open-source software. State media reported last month that both DeepSeek and Shanghai-based MiniMax had completed domestic registrations to open-source their latest models. Premier Li Qiang told the opening ceremony that China would pursue "open innovation" and promote technology transfer.

This year's WIC ran under the theme "Open Source and Openness," with about 1,800 delegates from 138 countries and regions, according to official figures. The AI panel was moderated by Wang Jian, an academician of the Chinese Academy of Engineering and founder of Alibaba Cloud; MiniMax's Wen Feng and the International AI Governance Association's John Higgins also took part. Alibaba owns the South China Morning Post.

What to watch next: how DeepSeek's model classification tiers actually work in practice, how the proposed export controls define a frontier model, and whether the open camp can agree on verifiable safety standards. The argument is already spreading beyond chatbots — at a separate WIC forum on Wednesday, Unitree Robotics founder Wang Xingxing said much of the data robots collect goes unused, and that self-improving physical AI models could put it to better work.