cross-posted from: https://lemmy.sdf.org/post/30359615
China’s ambassador to Australia has warned that a decision to ban artificial intelligence app DeepSeek from government systems and devices risks further politicising trade and technology ties.
[…]
Ambassador Xiao Qian’s comments came as a Chinese naval task force continued to skirt Australia’s territorial waters in an apparent plan to circumnavigate the island nation. The warships 10 days ago held live-fire drills in the Tasman Sea between Australia and New Zealand.
Writing in The Australian newspaper on March 3, Mr Xiao said the Chinese-developed AI program would “greatly benefit the world in various aspects” and encouraged Australia to work with Beijing to jointly develop new technologies.
“Taking restrictive measures against it under the pretext of ‘security risks’ is an attempt to overstretch the concept of national security and politicise trade and tech issues,” the ambassador said in his article.
In early February, Australia’s center-left Labour government became one of the first countries in the world to ban DeepSeek from official devices, a decision that it justified on national security grounds.
[…]
Deepseek isn’t open source.
Their models and a lot of their training and code to run it are
It is not open source according to the accepted definition. There is no such thing as “a lot” open source, or partly open source. This is just part of Deepseek’s PR campaign very much as many other false claims.
The code they have released is under the MIT licence which is most definitely an OSI approved Open Source licence.
The model’s licence grants rights to use and redistribute but imposes a number of conditions on usage. I would concede that it does not satisfy the conditions to be considered an open source license due to those conditions which exclude military use, harming minors, defamatory content, generating misinformation etc. Those restrictions might not be a problem for everyone but it isn’t Open Source. Meta’s LLama which is sometimes also claimed to be Open Source fails under similar criteria however is still popular with people running offline models on their own hardware.
The guys at Hugging Face have been working on an open reproduction of DeepSeek-R1, although I don’t know how they circumvent Chinese censorship (it’s also censored if you use Deepseek locally).
Another open source AI model beats DeepSeek with 86% less data.
All this Deepseek hysteria is just based on a simple press statement released by the company. It’s another totally over-hyped model with false claims that comes with even more disadvantages than most of its rivals.
If we’re going to be specific, then those restrictions just mean it’s not FOSS according to the Free Software Foundation. The source is still open, it’s auditable.
The model, which to most people is the far more important part, is not open source according to the criteria set by the Open Source Initiative. They own the trademark and police the Open Source® definition. It is fairly clear.
DeepSeek’s list of restrictions on use of their model puts them in a similar position to Meta’s LLama License is still not Open Source. I don’t think it makes sense to say a binary blob is either auditable or is source code but you can say the same of any LLM. There is no way to check the provenance or replicate it or re-build any of them.
The code they released is under an approved Open Source licence. The MIT licence is very permissive and is compatible with and can be incorporated into closed source and free software while being neither itself. No argument there.