OpenAI creates its own kryptonite!!

OpenAI, the company behind DALL-E and ChatGPT, has released its own kryptonite called AI Text Classifier. It is a GPT-3 and ChatGPT detector. OpenAI says it is meant to “distinguish between text written by a human and text written by AI.”

This new tool was trained using data from 34 different text generators across 5 companies, including OpenAI. This text was coupled with human-written content that was similar but not identical from sources including Wikipedia, websites scraped from Reddit links, and a set of “human demonstrations” collected for an earlier OpenAI text-generating system.

The AI Text Classifier has the potential to put a halt to the automated spread of incorrect information, plagiarism, and chatbots pretending to be human. But more progress is needed to reach these objectives. Classification accuracy is low at the moment. It has a 26% success rate, which is not good enough. On the other hand, it misidentified 9% of human-written pieces as being machine-made. OpenAI claims that the technology improves with longer content. When applied to brief texts, the classifier performs very poorly (below 1,000 characters), and it is not rare for the classifier to make a mistake when labelling even longer texts. Thus, it needs at least 1,000 characters(which is approximately 200-250 words) to do an analysis.

One place OpenAI is really focusing on with this detection tool is education. Its press release says that “identifying AI-written text has been an important point of discussion among educators,” as different schools have reacted to ChatGPT by?banning?or?blocking?it.

#ai?#chatgpt?