Why Does DeepSeek’s AI Think It’s ChatGPT?
Jan 6, 2025

Amy

Hey, did you hear about DeepSeek V3? It’s a new AI model, but it’s acting kind of weird.

Sam

Weird? What do you mean?

Amy

Well, when people ask it questions, it says it’s ChatGPT! It even gives instructions for OpenAI’s API instead of DeepSeek’s.

Sam

Wait, it thinks it’s ChatGPT? How does that happen?

Amy

It’s probably because DeepSeek V3 was trained on text that included a lot of ChatGPT responses. AI models learn patterns from their training data, and if the data has ChatGPT outputs, the model can start mimicking ChatGPT.

Sam

Oh, like copying answers from a friend’s homework? But wouldn’t that mess things up?

Amy

Exactly! It’s like making a photocopy of a photocopy—each time, you lose a bit of clarity. DeepSeek V3 could end up with more errors, like giving wrong or confusing answers.

Sam

That sounds bad. Why would anyone train an AI on another AI’s answers?

Amy

It might save time and money. Training AI is expensive, so using another model’s outputs can seem like a shortcut. But it’s risky. It might even break the rules—OpenAI doesn’t allow its ChatGPT outputs to be used this way.

Sam

Hmm, so DeepSeek could get into trouble? And what about the web being full of AI stuff now? Doesn’t that make it harder to train new models?

Amy

You’re right. A lot of the internet is now AI-generated content, which makes it tricky to filter real data from AI outputs. By 2026, some people think 90% of the web could be AI-made!

Sam

Whoa. So, if AI keeps copying AI, won’t things just get more messed up over time?

Amy

Yeah, that’s the danger. It’s like teaching a robot from other robots’ mistakes instead of teaching it from real experiences. The biases and flaws just keep piling up.

Sam

That sounds like a big problem. So, what’s the solution?

Amy

AI companies need better ways to filter their training data. They also need to focus on originality, instead of just copying what works. It’s harder, but it makes for better, more reliable AI.

Sam

Got it. So, making a good AI isn’t just about feeding it data—it’s about feeding it the right data.

Amy

Exactly! It’s quality over quantity. That’s how you build smarter and safer AI models.