Updated June 7, 2023 at 2:21 PM EDT
top line
A new study finds that an AI detector developed by the University of Kansas can detect AI-generated content in academic papers with 99% accuracy. This is one of the only detectors on the market dedicated to academic papers.
important facts
According to a report published in the journal on Wednesday, cell report physical science, Researchers have created a tool that can prove AI detection in academic papers with 99% accuracy.
A team of researchers selected 64 perspectives (types of articles) and used ChatGPT to create 128 articles using them to train an AI detector.
The model was 100% accurate in identifying human and AI-generated articles, and 92% accurate in identifying specific paragraphs in text.
tangent
A Study.com survey found that 89% of college students admit to using ChatGPT to help with assignments, and 34% of educators believe the software should be banned. , 66% support student access.
important quotes
“At the moment, there are some pretty obvious problems with AI writing,” lead author Heather Desaire said in a statement. It’s collected and there’s no such thing as a check for correctness. It’s like a game of two truths and one lie.”
Contra
No AI detector has been proven to be 100% accurate. A student at the University of California, Davis, claimed he was wrongfully accused by the university of cheating with AI. After uploading a paper in one of her classes, she received an email from a professor who claimed she was partially flagged as AI-generated in her program Turnitin. Her case was immediately transferred to the Office of Student Aid and Justice, which is responsible for disciplining academic misconduct. The student filed a lawsuit and eventually won by proving that he used the timestamp to write the paper. This wasn’t the only time educators were mislabeling assignments as AI-generated. A commerce professor at A&M, Texas, tried to fail more than half of his advanced class after using ChatGPT to test whether students were writing papers using chatbots. He copied and pasted the document into his ChatGPT and asked if it was written and he said yes. Because of this, the professor offered them the opportunity to redo their assignments, but the university withheld their diplomas. However, the university confirmed to insiders that no students were rejected or prevented from graduating.
main background
OpenAI’s chatbot, ChatGPT, was released to the public in November 2022, and in less than a week it had over 1 million users, people using it to code, write essays, and more. AI intelligence has led several schools, including New York Public Schools, Seattle Public Schools, and the Los Angeles Unified School District, to ban software indefinitely or temporarily.
Tangent 2
There are already many existing programs and services that promise to identify AI-written content.
- turn it in Released an AI detection tool for papers. Previously, it only had the ability to check for plagiarism. This feature has been added to the Similarity Report to show the overall percentage of work generated by the AI software in your paper. The company claims its AI detection tools are 98% accurate in detecting AI-written works. Its detection model is trained to detect content from his GPT-3 and GPT-3.5 language models, including ChatGPT. However, GPT-4’s write characteristics are similar to previous models, so it claims to be able to “mostly” detect content in this version.
- copy leak The detection accuracy of AI is said to be 99%. The company’s software can detect AI-generated text across multiple models, including versions prior to GPT-4 and content created with Jasper AI. It also says it can detect AI content in multiple languages, including Spanish, Russian, French, Dutch, and German.
- Winston AI It launched in February and claims to perform this task with 99% accuracy. Detection is only supported in English and French, but the company plans to expand this to Spanish and German soon. It can detect content created using ChatGPT, Bard, Bing Chat, GPT-4, and other text generation tools.
- OpenAI classifier It was launched in January to distinguish AI-written text from human-written text. These are from the same makers as ChatGPT, but this tool is not very accurate. It has a success rate of about 26%, with a 9% chance of misclassifying human work as AI work. However, OpenAI claims that accuracy improves as text length also increases. OpenAI recommends using the software only for documents written in English, as it is “extremely unreliable” for text under 1,000 characters. The company also cautions against using Classifier as a primary decision-making tool and suggests using Classifier to complement other detection methods.
- AI writing check Developed by Quill and CommonLit, it allows teachers to see AI-generated work in assignments. The developer estimates its accuracy to be between 80% and 90%. Only up to 400 words of text can be detected at a time, otherwise the user must divide the text into sections. This detection software, created by her OpenAI, can identify non-human linguistic syntactic patterns in text.
References
ChatGPT in Schools: Where It’s Banned and How It May Affect Students (Forbes)
What You Need to Know About OpenAI’s ChatGPT — What It Breaks and How It Uses (Forbes)
follow me twitter Or LinkedIn. Send me your safety tips.
