Summary
OpenAI claims Chinese AI startup DeepSeek used its proprietary models without authorization, allegedly extracting large amounts of data via OpenAI’s API.
The company blocked suspected accounts for violating its terms of service.
The revelation comes as OpenAI restructures into a for-profit entity and deepens its partnership with Microsoft.
DeepSeek’s emergence has disrupted AI markets, briefly affecting NVIDIA’s stock.
Lmao that’s rich coming from them of all people
Pot calling the kettle black…
… yep, they’re rich. This saying has reached a full circle.
In Portuguese we have an expression that would roughly translate to “a thief who steals from a thief is forgiven for a thousand years”.
I don’t remember it ever applying so literally.
The same expression exists verbatim also in Spanish: “Ladrón que roba a ladrón tiene 100 años de perdón”.
I’m still waiting for my check for contributing nonconsenually to openAI training.
Just wait until they notice that the data that OpenAI had harvested from the net was illegally obtained from the start…
This is what my grandma meant when she said “this is the pot calling the kettle black”
Yes, OpenAI wishes everyone else has to have authorization to do model training…
Fortunately, their ToS don’t matter all that much, it’s easy to use their model through a third party without ever touching them.
It’s afraid
So many are bitching about how, “there’s no way it only cost $5.6 million (to train DeepSeek).” Yes, yes there is: It’s called subsidies and government assistance.
What do I mean? A simple example we can all understand: When you buy something from somewhere in China (e.g. AliExpress) the shipping price is often free or so cheap as to be negligible. How can it be that shipping is free on a $1 set of earrings‽ It’s because China subsidizes the shipping!
China isn’t really a capitalist nation (they’re not communist either but that’s a different matter). That means they have no qualms whatsoever about giving businesses resources for free or next to nothing. That includes loans, reduced shipping, employee recruitment (and straight up loaning them experts), and AI training data.
Did the folks who made DeepSeek copy trade secrets from OpenAI? Sure, that’s possible but why would they need to? China has access to the same free-on-the-Internet training data as everyone else plus a very, very special resource no else has: Everything that’s ever gone through WeChat.
That’s another thing about China: If they come into your business and tell you, “give your data to this other business” you have no say in the matter. You bow and hand it over or you get arrested and maybe don’t live for much longer.
This sucks for businesses 99% of the time but every now and again it gives China a massive competitive advantage when everything comes together juuuust right. This is one of those cases.
China isn’t really a capitalist nation
They’re a mixed state-capitalist and private-capitalist country. Also a totalitarian dictatorship with a propensity to commit genocide against its minorities.
I’m curious to see in a few months time if independent researchers are able to validate or refute the claim that it took only $5 million to train the model.
