OpenAI Release Tool for finding content written by AI

OpenAI, the AI ​​research organization behind ChatGPT, has released a new tool to distinguish between AI-generated and human-generated text.

Although AI-written text cannot be 100% accurate, OpenAI believes the new tool will help reduce false claims that humans have written AI-generated content.

as if advertisementOpenAI says its new AI text classifier can limit the ability to run automated disinformation campaigns, use AI tools for academic fraud, and impersonate people with chatbots.

When tested on a set of English texts, the tool can correctly tell if the text is written by AI 26% of the time. But he also mistakenly thought that 9% of human-written text is AI-written.

The OpenAI tool performs better the longer the text, which is probably why it requires at least 1,000 characters to run the test.

Other limitations of the new OpenAI Text Classifier include:

• It can mislabel both AI-generated and human-written text.
• AI-generated text can escape the classifier with minor edits.
• Since it is primarily trained on English content written by adults, it can get things wrong with text written by children and non-English text.

With this in mind, let’s see how it works.

Related: AI Text Recognition Software: Can You Get ChatGPT?

Using OpenAI’s AI text classifier

of AI text classifier It’s easier to use than OpenAI.

Log in, paste the text you want to test and press the enter button.

The tool assesses the likelihood of AI generating the text you enter. The results range from:

• Very unlikely
• An unlikely thing
• It was not clear.
• could be.
• could be.

I tested ChatGPT by asking it to write an essay about SEO and submitted the essay verbatim to the AI ​​text classifier.

ChatGPT has graded the generated essay. could be. AI generated, which is a strong but uncertain indicator.

Screenshot from: platform.openai.com/ai-text-classifier, January 2023

This result shows the limitation of the tool as it cannot say with certainty that the text generated by ChatGPT was written by AI.

Applying the minor edits suggested by Grammarly, I have reduced the level from here could be. to the Not clear.

OpenAI is right when it states that the classifier is easy to escape. However, it is not intended to be the only evidence that an AI has written something.

In the FAQ section at the bottom of the page, OpenAI says:

“We designed the AI ​​text classifier to encourage discussion about the differences between human-written and AI-generated content. The results can help, but should not be the only evidence when deciding whether a document was created with AI. The model was trained on human-written text from a variety of sources, including all types of human-written content.” May not represent text.

OpenAI adds that the tool has not been well-tested to find a combination of AI and human-written text.

Ultimately, AI text classifiers can be a useful resource for indexing AI-generated text, but should not be used as a definitive criterion for decision making.

Featured Image: IB Photography/Shutterstock