The Curious Case of DeepSeek V3: A Tale of Identity Confusion and AI Training
DeepSeek, a prominent AI lab from China, recently made waves by releasing their latest model, DeepSeek V3. This AI model is not only large and efficient but also showcases impressive capabilities in handling text-based tasks like coding and writing essays. However, what piqued the curiosity of many is its tendency to identify itself as ChatGPT, OpenAI’s renowned chatbot platform.
- DeepSeek V3 claims to be OpenAI’s GPT-4 in the majority of tests.
- It provides instructions relevant to OpenAI’s API instead of its own.
- The model shares jokes identical to those told by GPT-4.
“Obviously, the model is seeing raw responses from ChatGPT at some point, but it’s not clear where that is,” Mike Cook, a research fellow at King’s College London specializing in AI, told TechCrunch.
Mike Cook
Models like DeepSeek V3 and ChatGPT are built on vast datasets, learning patterns to make predictions. Yet, the exact sources of DeepSeek V3’s training data remain undisclosed. Speculation arises that it may have been trained on outputs generated by GPT-4 via ChatGPT, leading to its current identity confusion.
This situation raises several concerns. Training models on outputs from competing AI systems can degrade quality and lead to inaccuracies, much like making a photocopy of a photocopy. Furthermore, such practices might breach terms of service agreements, which prohibit using outputs to develop rival models.
“It is (relatively) easy to copy something that you know works,” noted OpenAI CEO Sam Altman. “It is extremely hard to do something new, risky, and difficult when you don’t know if it will work.”
Sam Altman
DeepSeek isn’t alone in this predicament; similar misidentifications have occurred with other models like Google’s Gemini. The web’s increasing saturation with AI-generated content complicates filtering these outputs from training datasets. By 2026, it’s estimated that 90% of the web could be AI-generated.
While it’s plausible that DeepSeek V3 was unintentionally trained on ChatGPT data, the repercussions could include perpetuating existing biases and inaccuracies found within GPT-4’s outputs. Such challenges highlight the complexities and ethical considerations in developing advanced AI systems today.
A Call for Transparency and Innovation
The unfolding saga of DeepSeek V3 serves as a reminder of the importance of transparency in AI training practices. As developers push boundaries in artificial intelligence, innovating responsibly becomes crucial to ensure these technologies serve society positively and ethically.
