The company has previously attempted to detect AI-generated text but shut down its previous AI text detector due to low accuracy. The new approach focuses solely on detecting writing from ChatGPT, not from other companies’ models. Despite its high accuracy and effectiveness against localized tampering, the method is less robust against globalized tampering and could be easily circumvented.
Key takeaways:
- OpenAI has developed a tool that could potentially detect students who use ChatGPT to write their assignments, but the company is still debating whether to release it.
- The tool uses a text watermarking method, which involves making small changes to how ChatGPT selects words, creating an invisible watermark that can be detected later.
- Despite its potential, the method has its limitations, including susceptibility to circumvention by bad actors and the potential to disproportionately impact non-English speakers.
- OpenAI had previously shut down its AI text detector due to its low accuracy rate, and the new text watermarking method represents a different approach to detecting AI-generated text.