The AI chatbot developed by OpenAI, ChatGPT, has gained widespread popularity among internet users who seek answers to diverse queries and tasks within seconds.
However, this trend has caused serious concerns among educators who are worried about students submitting papers written by ChatGPT.
In response, researchers at Stanford University have created a tool called DetectGPT to help teachers identify content generated by ChatGPT and other similar large language models (LLMs).
After New York prohibited students at public schools from utilizing ChatGPT for academic purposes, OpenAI has confirmed that it is in the process of creating a detection tool.
DetectGPT is a tool created to identify texts written by ChatGPT or similar chatbots. The researchers behind it say that these AI-generated texts “occupy negative curvature regions of the model’s log probability function.”
This information helps DetectGPT decide if a text was written by a specific language model by checking the log probability. The researchers claim that DetectGPT is correct 95% of the time.
A log probability function is a mathematical concept used in the field of AI and machine learning. It is used to determine the probability or confidence level of an event happening. The logarithmic transformation of probabilities helps make it easier to compare and interpret the results, especially in applications such as natural language processing and language generation. This allows for a more accurate evaluation of the text generated by the AI models.
Educational institutions are worried that students are using tools like ChatGPT to write their papers and homework, so they created a tool called DetectGPT to help catch them.
Unfortunately, the team behind DetectGPT has not yet disclosed much information about their tool, and it is not currently accessible to the general public.
Nevertheless, the development of such tools is a step in the right direction toward ensuring that students complete their work in an honest and trustworthy way.