On Tuesday, AI research startup, OpenAI unveiled a tool to check ChatGPT and other AI tools. The new tool can if a text comes from an AI tool or from a human. After it was released two months ago, ChatGPT went viral. This is a chatbot that can generate human-like prose from prompts. Microsoft revealed last week that it had invested several billion dollars in the startup OpenAI. It also claims that it would use the startup’s AI models in its coming products. There were several worries that students may use the AI tool to cheat or copy. For this reason, some of the biggest school districts in the U.S., including New York City, have banned ChatGPT.
Gizchina News of the week
OpenAI’s new tool is a work in progress
“In our evaluation on our ‘challenge set’ of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as ‘likely to have been written by an AI’, while human-written 9% of the text was incorrectly labelled as written by AI (false positives),”.
For articles with less than 1000 letters, the new tool is unreliable, and OpenAI does not advise using it in languages other than English. In addition, text written by AI can be altered slightly, thereby making it impossible for the classifier to correctly determine that it is not the product of a person.
OpenAI says it is contacting educators to talk about ChatGPT’s strengths and weaknesses and will keep working on identifying AI-generated material. OpenAI made it obvious as early as 2019 that detecting text created by artificial intelligence is not a simple task. An employee of OpenAI stated, “Our work on detecting AI-generated text continues, and we aim to provide improved methods in the future.”