DeepSeek engineer says open source is a race against time, and admits it adds safety risk
At China's World Internet Conference, a DeepSeek engineer explained why the lab publishes its work and conceded that openness can magnify AI's safety risks.
The short version
- Gao Xi spoke on 3 November at the Wuzhen summit's AI dialogue, saying DeepSeek open-sources work to speed up AGI.
- He argued closed models may keep a short-term lead but delay everyone's access to the benefits of the technology.
- He admitted open source can amplify AI safety risks and said DeepSeek has a duty to prevent that.
- Alibaba's Qwen lead cited 460+ open models and 40亿 downloads; MiniMax called open models more controllable.
- Proposed Chinese export rules would require reporting before open-sourcing certain restricted technologies.
Why it matters
This is a rare first-person explanation of why a leading open-weight lab publishes its work. It suggests openness is a deliberate strategy, not a favor. It is still a speech, not a release commitment, so check the license on each model you actually use.
Why DeepSeek gives its work away
As BAAI Community reported, DeepSeek research engineer Gao Xi took the stage at the World Internet Conference in Wuzhen on 3 November and answered a question many people ask: why publish hard-won research? His answer was that it serves the goal of AGI, short for artificial general intelligence, meaning AI that can match people across most tasks.
His framing of the race was blunt: "The opponent is not other companies, but time." He gave no forecast for when AGI arrives. He only said he would like it sooner, whether it was due in a dozen years or in one.
The logic of sharing early
Gao's example was a new method, such as a cheaper way to train a model. The fastest path, he said, is to publish it so others can discuss and test it. If someone else solves the same problem, DeepSeek can adopt their result quickly; if DeepSeek solves it first, sharing moves everyone forward.
Closing off work to block rivals might hold a lead for a while. But over a longer period, he argued, it only delays when society as a whole gets the benefits.
Admitting the risk
Gao also said open source can indeed magnify the safety risks of AI, and that DeepSeek is responsible for stopping that from happening. According to the reports, he mentioned grading models by risk but gave no details.
Other panelists went further. Alibaba's Qwen representative said open models can't be recalled once released and lack outside guardrails, since third parties can bypass built-in protections. He still argued for broader governance, including laws and industry standards, rather than limits on the models alone. MiniMax's representative said open models give users more control, because they can fine-tune and host them privately.
The rules in the background
On 16 October, Chinese ministries opened a consultation on revising the export-restricted technology list, with 23 of 40 proposed new items related to AI. Technology already released openly would fall outside the rules, but newly open-sourcing a restricted technology would require reporting first. The comment period ends on 15 November.
Reports say DeepSeek and MiniMax have already completed the related registration. Whether the final list widens or narrows the path for open releases is not yet known.
What to watch
- Whether the final export-control list, after the 15 November comment deadline, makes open releases easier or harder.
- Whether DeepSeek publishes details of the risk grading Gao mentioned.
Try it
If you build on open-weight models, read the license and model card for each release you depend on, instead of inferring plans from speeches.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by BAAI Community · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
