In a move that makes AI detector tools redundant, OpenAI yesterday announced that it has trained a classifier, alongside launching AI Text Classifier to distinguish between text written by a human and text written by AIs from a variety of providers.
Citing automated misinformation campaigns, OpenAI said while it is impossible to reliably detect all AI-written text, it believes that classifiers can inform mitigations for false claims that AI-generated text was written by a human.
Compared to its previously released classifier, this new classifier by OpenAI is significantly more reliable on text from more recent AI systems. The company has made this publicly available to get feedback, checking if tools like this are useful.
Check out the AI Text Classifier here.
AI Text Classifier
AI Text Classifier is a fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, including ChatGPT and alike. OpenAI said that this classifier is available as a tool to spark discussions on AI literacy.
Limitations of AI Text Classifier
- A minimum of 1,000 characters are needed for the classifier to run, which is roughly 150–250 words.
- The classifier isn’t always accurate; it’s capable of mislabeling AI as well as human-written texts.
- Editing AI-generated content is simple and can be used to manipulate the classifier.
- Since it was mostly trained on English content created by adults, the classifier is likely to make mistakes when dealing with text written by non-English texts and kids.
- Text that is very universal and predictable cannot be reliably identified like the first five alphabets will be the same.
- Neural network-based classifiers perform poorly when used outside of training data. OpenAI said that the classifier can occasionally be very certain that a false prediction will be made for inputs that are substantially different to the text in our training set.
We tried the AI Classifier with an essay generated by ChatGPT and here is the result.
The classifier does not assert that the text is generated by AI but agrees that it ‘possibly‘ is.
ChatGPT Ban
This comes in the backdrop of multiple schools and educational institutions raising concerns about students using ChatGPT for answers. ChatGPT has drawn flacks of educational institutions across the world as it can easily write essays, codes, scripts, and poetry and prose of all kinds. Considering the backlash, OpenAI has developed a preliminary resource on the use of ChatGPT for educators, which outlines some of the uses and associated limitations and considerations. OpenAI is working with US schools to find solutions for the problem.