OpenAI, the maker of the popular chatbot ChatGPT, has released a new software tool that aims to identify text generated by AI. The AI classifier, trained on a human and AI-written text dataset, is designed to address issues such as academic dishonesty and automated misinformation campaigns. However, the detection tool is still in its public beta mode. OpenAI acknowledges that it can be unreliable on texts under 1,000 characters and can be easily tricked by editing AI-written text. The company seeks feedback from educators and parents to improve the tool's accuracy.
ALSO READ: ChatGTP Plus launched in the US as a new subscription service
OpenAI's latest tool is more advanced than its previous version from 2019 but still has limitations. The classifier correctly identifies 26% of AI-written text and mislabels human-written text as AI-written 9% of the time. The tool is not very effective on texts under 1,000 characters and should not be used for languages other than English. Despite these limitations, OpenAI says it will continue to work on detecting AI-generated text and hopes to improve the method.
ALSO READ: ChatGPT becomes the fastest-growing app in the world, records 100 million users in 2 months
The release of OpenAI's new tool follows the attention garnered by ChatGPT and Microsoft's recent multi-billion dollar investment in OpenAI. The widespread popularity of ChatGPT has led some school districts in the US, including New York City, to ban the chatbot over concerns that students will use it to cheat or plagiarize.
1. What is the AI classifier tool released by OpenAI?The AI classifier is a software tool developed by OpenAI to distinguish between text written by humans and text generated by AI. The tool uses machine learning to identify the likelihood of a text being written by AI and is intended to address issues like misinformation campaigns and academic dishonesty.
2. How reliable is the AI classifier tool?
According to OpenAI, the AI classifier tool is still in its public beta mode. It is not very reliable for texts under 1,000 characters. Additionally, AI-written text can be edited to trick the classifier, so it is not a foolproof solution.