OpenAI develops watermarking method for AI-generated text, but concerns delay release.

OpenAI, the creator of AI chatbot ChatGPT, has developed a watermarking method to identify AI-generated text but has yet to release it due to internal disagreements and concerns over potential negative impacts. The watermarking system modifies ChatGPT's word prediction to create a unique pattern, but some employees worry it may deter users and be easily circumvented. OpenAI is considering alternative methods, such as embedding metadata, to address these issues.

August 04, 2024
7 Articles

Further Reading