It Looks Like DeepSeek Took TOO MUCH Inspiration From ChatGPT

DeepSeek came in like a wrecking ball. Before the company’s arrival, many assumed OpenAI’s ChatGPT would continue to dominate the AI scene. However, DeepSeek’s capabilities and the reports that the company only spent $6 million training the model was what turned the AI industry on its head. However, a new Copyleaks report has revealed that maybe DeepSeek was so effective simply because it trained itself on its chief competitor—OpenAI.

Sharing a similar style

Copyleaks, for those unfamiliar, is a company that specializes in AI-based text analysis, AI governance, and plagiarism detection. According to its recent findings, it discovered that 74.2% of texts generated by DeepSeek-R1 match OpenAI’s stylistic fingerprints. This means that there is a very good chance that DeepSeek trained itself on OpenAI’s outputs.

According to the company, its research used a combination of three advanced AI classifiers. It trained each of these classifiers on texts from four major AI models: Claude, Gemini, Llama, and OpenAI. The classifiers help to identify subtle differences between the AI models. This includes how sentences are structured, choice of vocabulary, and phrasing.

Based on this, DeepSeek’s style of writing matched OpenAI’s 74.2%. Should we be surprised? No. For those unfamiliar, DeepSeek claimed that it saved a lot of money by using a training method called distillation. Instead of training an AI model from scratch, distillation involves taking the output from already-trained AI models like ChatGPT and using that to train itself.

Think of it as the relationship between the student and the teacher. The teacher might have had to learn about a topic from scratch, doing their own research and experiments to arrive at a conclusion. The teacher then takes this knowledge and distills the correct information to the student. So, instead of the student spending years learning about a subject, they could easily grasp the knowledge in a single semester (we’re talking about competence, not mastery).

Troubling findings

According to Copyleaks, this discovery is worrying. If you’ve been following the news, when DeepSeek arrived, it blew a trillion-dollar hole in the US stock market. This is because investors bought into the story that companies didn’t need to spend billions to train AI. This resulted in share prices of companies like NVIDIA, which make and sell hardware used in AI development, to plummet.

Copyleaks suggests that based on these findings, DeepSeek might have misled the market and given it an unfair advantage. We did highlight this in our DeepSeek feature story, where the company hasn’t been forthcoming regarding its training data. Since we don’t know how the model was trained, it begs the question of legitimacy, where we have to wonder if we can trust the answers it gives.