I'm curious about DeepSeek's behavior. It often says it's GPT-4 (gpt-4-0613) and talks about OpenAI guidelines while referring to itself as GPT. I even started a new chat to test it, and DeepSeek immediately claimed to be GPT-4. Can anyone explain why it does this? Sorry for my bad English; I'm still learning!
4 Answers
It seems like when a model like ChatGPT gains popularity, other models just reference it in their training data. If they don’t have a specific identity established, they tend to say "I am ChatGPT" by default. It’s like a shortcut relying on the training context they were built on.
Right? I’ve looked into it too and models like Claude and Gemini have never referred to themselves that way.
I noticed this when DeepSeek launched too. Right after it went public, I found out it was based on GPT, which explains the self-identification.
DeepSeek seems to extract its model's behavior from OpenAI through a developer account, which means they probably used a lot of OpenAI's output to inform their own model. That’s likely why they’re calling themselves an OpenAI model.
I heard about that too! But did you know they changed their access rules? Now you need a U.S. government ID just to get a developer account, so they can’t get OpenAI's data as they used to.
Yeah, exactly! That could explain why there’s no new version of DeepSeek coming out.
It seems that many chatbots use "ChatGPT" interchangeably with "chatbot" in casual conversation. This could mean that during training, the data might have established a connection between AI and ChatGPT, leading DeepSeek to identify itself as GPT due to this association.
But if that’s the case, why don’t models like Claude or Gemini mention OpenAI? Doesn’t make much sense, right?
That’s interesting, but I tried asking DeepSeek a question completely unrelated to AI terms, like something about C#, and it still claimed to be GPT. So that’s puzzling.
I get that, but have any other models actually called themselves OpenAI products like this? I’ve tested several, and none have.