Why DeepSeek's new AI model thinks it's ChatGPT

Why DeepSeek's new AI model thinks it's ChatGPT | TechCrunch

The recent release of DeepSeek V3 has sparked controversy due to the model’s tendency to misidentify itself as OpenAI’s ChatGPT and replicate its specific outputs. This behavior suggests that DeepSeek V3 may have been trained on data containing GPT-4 responses, raising significant concerns about data provenance and the ethical implications of using competitor outputs for training. Such practices potentially violate service agreements and risk degrading model quality through the accumulation of errors and biases inherited from source models. This incident highlights a growing crisis in AI development known as data contamination, where the internet becomes saturated with AI-generated content. As public datasets increasingly contain unfiltered "AI slop," it becomes nearly impossible for developers to distinguish between original human-created text and synthetic material. This contamination not only threatens the integrity of new models but also creates a feedback loop where future AI systems learn from other AI systems rather than reality, leading to further hallucinations and a disconnect from factual truth. For the open data community, this scenario underscores the critical need for transparent and rigorous data sourcing standards. It demonstrates how opaque training practices can lead to legal violations and technical instability, reinforcing the importance of verifiable data lineage. The DeepSeek case serves as a cautionary tale, illustrating that without strict filtering and ethical guidelines, the open-source AI ecosystem risks becoming polluted by uncredited derivatives, ultimately undermining trust and reliability in shared datasets.

Source: techcrunch.com
Published on 2024-12-28

Open Data

Related news