this post was submitted on 14 Oct 2025
99 points (100.0% liked)
memes
23628 readers
270 users here now
dank memes
Rules:
-
All posts must be memes and follow a general meme setup.
-
No unedited webcomics.
-
Someone saying something funny or cringe on twitter/tumblr/reddit/etc. is not a meme. Post that stuff in /c/slop
-
Va*sh posting is haram and will be removed.
-
Follow the code of conduct.
-
Tag OC at the end of your title and we'll probably pin it for a while if we see it.
-
Recent reposts might be removed.
-
No anti-natalism memes. See: Eco-fascism Primer
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Allegedly they used API calls to OpenAI to train the model initially. And that's basically no different from the scraping done by all the AI companies.
https://www.nbcnews.com/tech/tech-news/openai-says-deepseek-may-inapproriately-used-data-rcna189872
allegedly. According to OpenAI.
The thing is, DeepSeek is only relevant because its R1 model was the first in the world to match openAI's o1 model on reasoning. This was at the time openAI was thought to have a huge lead there. Here's the thing: OpenAI specifically blocked every way to access the reasoning traces from o1 via the API and their apps. Deepseek invented their own algorithm to generating reasoning traces, and open sourced it, and now every AI lab uses it and can compete with OpenAI.
The claim that Deepseek copied them doesn't hold water: because HOW, when they blocked all access to reasoning specifically to prevent that?
And if the argument is: they just copied the answers, that is a lame one because there's whole public datasets built off chatgpt answers that the open source community has been using for years; that's a nothingburger story and does not explain why deepseek was so good. Remember for 6 months, not a single lab was even close to matching o1 until deepseek showed the world how. And then suddenly everyone could.
I think that OpenAI put that out there to feed into the 'china copycat' narrative to fearmonger USA legislators into harsher hobbling of the chinese AI industry via blocking GPU exports. OpenAI pushed for this, Anthropic pushed for this, Google pushed for this.
I'm well aware, DeepSeek took an innovative approach using mixture of agents and reinforcement learning to drastically reduce the cost of running models. They're continuing to push the envelope incidentally, and got the best paper at ACL just a few months ago.
The claim was that DeepSeek used the existing API for ChatGPT to do the initial tuning of the weights of their model. That's entirely possible. And that's the whole irony which makes the meme funny. OpenAI has some EULA on how you're allowed to use the API, and they claim DeepSeek violated that. However, they themselves have never had a problem siphoning data from any source they could find without any regard for permissions from the authors.
Obviously, OpenAI is terrified of DeepSeek because it undermines their whole business model. Everybody would prefer using an open model they can tune any way they like and run on premises without having to send their data to another company. So, OpenAI is very obviously smearing DeepSeek any way they can. It's just funny that the smears themselves expose their own hypocrisy.