OpenAI launches tool for detecting AI-generated text

OpenAI LLC immediately launched a cloud-based instrument that may assist customers decide whether or not a chunk of textual content was written by synthetic intelligence.

The instrument, which is offered at no cost, is itself based mostly on an AI mannequin. It may detect textual content generated by not solely OpenAI’s neural networks but additionally competing software program from different corporations. Nevertheless, the startup cautioned that the instrument presently has numerous vital accuracy limitations.

“In our evaluations on a ‘problem set’ of English texts, our classifier appropriately identifies 26% of AI-written textual content (true positives) as ‘probably AI-written,’ whereas incorrectly labeling human-written textual content as AI-written 9% of the time (false positives),” OpenAI researchers detailed in a blog post immediately.

OpenAI says that the instrument is most adept at analyzing English-language texts comprising greater than 1,000 characters. It may’t reliably consider textual content snippets which can be shorter or written in different languages. Furthermore, the instrument will not be but able to precisely detecting AI-generated software program code.

There are additionally different limitations to the instrument’s accuracy. In response to OpenAI, AI-generated textual content might be modified to make detection harder. The startup’s researchers added that “for inputs which can be very totally different from textual content in our coaching set, the classifier is typically extraordinarily assured in a mistaken prediction.”

The coaching dataset that OpenAI used to develop the instrument included a group of textual content prompts submitted by customers to its cloud-based AI fashions. In response to the corporate, its researchers entered every immediate into a number of neural networks to supply a sequence of machine-generated responses. The corporate additionally used textual content from different sources throughout the growth course of.

OpenAI’s efforts to develop software program for detecting AI-generated textual content date again a number of years. In 2020, the startup released GPT-2 Output Detector, a instrument for recognizing textual content generated by its GPT-2 neural community. The neural community is a predecessor of ChatGPT that was launched in 2019 and gives a extra restricted characteristic set.

In response to OpenAI, the brand new instrument it launched immediately is considerably more proficient at detecting textual content generated by not too long ago developed AI fashions.

“Whereas it’s inconceivable to reliably detect all AI-written textual content, we imagine good classifiers can inform mitigations for false claims that AI-generated textual content was written by a human: for instance, working automated misinformation campaigns, utilizing AI instruments for educational dishonesty, and positioning an AI chatbot as a human,” OpenAI’s researchers defined.

The discharge of the instrument comes per week after the startup raised a multibillion-dollar funding from Microsoft Corp. to advance its AI analysis. The funding consists of not solely capital but additionally cloud computing infrastructure. Beforehand, OpenAI raised $1 billion from Microsoft in 2019 and partnered with the corporate to construct a cloud-based supercomputer for its researchers. 

Picture: OpenAI