Has DeepSeek Been Using ChatGPT’s Data Without Permission? New Findings Could Shatter AI Trust!

According to a new study by Copyleaks, an AI detection firm, 74.2% of the texts written by DeepSeek are similar to ChatGPT which suggests that DeepSeek may have been trained on outputs from ChatGPT. Algorithm classifiers and screening technology were used by Copyleaks to look at different styles of texts written by ChatGPT, Gemini, DeepSeek, and Llama. When looking at all those texts, users will find that every AI model has its unique and distinguished style of text but DeepSeek’s outputs seem like they have been generated by ChatGPT.

The head of data science at Copyleaks, Shai Nisan, says that this raises alarming questions about how DeepSeek was trained and if it used ChatGPT’s outputs, it did so without authorization. The research only focuses on writing styles of AI models, and it's just like a handwriting expert matching different handwritings. Now that the data suggests that DeepSeek has a similar writing style to that of ChatGPT, proper investigation has to be made.

If it's true that DeepSeek has been using ChatGPT’s data for training, it would mean that DeepSeek has violated the intellectual property of ChatGPT. This also highlights that there is a need for regulatory frameworks that stop other AI models from using datasets.

This issue is more concerning if we look at how much DeepSeek has impacted the AI market, and there are a lot of questions surrounding its tech. If there is evidence which shows that DeepSeek has used outputs of ChatGPT unauthorized, it could make a significant impact on DeepSeek’s future. OpenAI is already facing criticism for training its AIs on web content without any proper permission, so this adds more complexity. There is an absence of legal standpoints for AI, so it can become an issue as well.

It also suggests that similarities between ChatGPT and DeepSeek can be because of deeper structural ot training similarities which overlapped and AI fingerprinting will be needed for further analysis. If there are more solid evidences found, further investigations will have to be done to decide the future of DeepSeek.

Image: Saradasish Pradhan / Unsplash

Read next: Google Processes 158,548 Searches Every Second, Here’s How Much It Adds Up to in Just One Year!
by Arooj Ahmed via Digital Information World