OpenAI Develops Tool to Detect ChatGPT Texts, Hesitates on Release

An American company has found that a technology operational for a year could harm their AI, suggesting that ChatGPT's market entry might lead to user loss. How can these risks be mitigated?

OpenAI: Innovative yet Cautious

Previously, detecting AI-generated text was challenging. OpenAI, a leader in AI technology, has paved the way by developing a system to identify texts produced by its flagship text generator, ChatGPT.

A Controversial Innovation

According to the Wall Street Journal, OpenAI developed this detection system about a year ago but has not yet deployed it.

The system’s approach is straightforward: it slightly alters ChatGPT’s word choices to create a specific identifiable pattern. However, this ChatGPT-specific detector does not work on texts from competitors.

Internal Debate Over Deployment

Whether to make this technology public has sparked internal debates at OpenAI for two years. A company spokesperson stated, “The method we are developing is technically promising, but also poses significant risks we are assessing.”

Non-native English speakers using ChatGPT as an assistant might be disproportionately affected.

Revelations and Updates

Following revelations by the Wall Street Journal, OpenAI provided additional details in a May blog post.

The company acknowledges that while the system effectively counters local paraphrasing, it is easily bypassed by malicious actors who extensively paraphrase an entire text.