OpenAI releases ChatGPT artificial intelligence text generation detection tool

​ChatGPT

ChatGPT is an interactive natural language dialogue model launched by OpenAI in November 2022. OpenAI uses RLHF (Reinforcement Learning from Human Feedback) to train ChatGPT, which is the same method used by InstructGPT, but there are obvious differences in the data collection settings. Use supervised tuning (parameter) when training the initial model: Sessions provided by a human AI trainer. Trainers have access to model suggestions to help the trainer complete responses better. The newly generated conversational dataset is then mixed with the InstructGPT dataset to form a conversational format.

ChatGPT caused a shock in the industry after its release, and was subsequently widely trialed by all parties, including students. Some students try ChatGPT to write papers, assignments and programs, and it is difficult for teachers to tell whether the papers, assignments and programs are completed by the students themselves. In addition, there have been many unpublished papers signed ChatGPT.

In response to this problem, schools around the world have taken measures, such as banning the platform. According to Business Insider, ChatGPT has been banned by middle and high schools in New York, Seattle, Los Angeles and other places, and the platform has also been banned by university computers in France and India.

ChatGPT detection tool

On January 31, OpenAI announced the launch of an artificial intelligence text classifier, designed to detect whether input content is generated by artificial intelligence tools such as ChatGPT.

OpenAI says its classifier is not completely reliable. When tested on English text, the classifier correctly identified 26% of the text generated by artificial intelligence, and about 9% of the content written by humans was incorrectly labeled as generated by artificial intelligence. The performance of the classifier improves as the input text length increases. Compared with other previous classifiers, the accuracy of this classifier is higher than other detection classifiers.

Classifier trial address: http://platform.openai.com/ai-text-classifier

OpenAI said that because the results are not completely reliable, it is not recommended to use this tool for decision-making , but as a complement to other detection tools and methods.

ChatGPT trial address: https://chat.openai.com/

https://openai.com/blog/new-ai-classifier-for-indicating-ai-written- text

This article is translated from: https://www.bleepingcomputer.com/news/technology/openai-releases-tool-to-detect-ai-written-text/

The above is the detailed content of OpenAI releases ChatGPT artificial intelligence text generation detection tool. For more information, please follow other related articles on the PHP Chinese website!