Pages

Wednesday, July 26, 2023

Nondetection

As we have noted in the past, efforts to detect cheating via artificial intelligence programs such as chatGPT often produce false positives, i.e., they indicate a document was written by AI when it wasn't. 

UCLA has discouraged use of such programs precisely because of the false positive danger, which  could result in a student being incorrectly charged with cheating.

Now, OpenAI has admitted that its detection program is useless. The notice below comes from its website:


For more information (but no solution), check out:

https://decrypt.co/149826/openai-quietly-shutters-its-ai-detection-tool.

There are clues that cheating might be occurring. Perfect grammar is one, although a smart student could stick in a few errors. Another clue is references that turn out to be invalid or information that you know to be incorrect - since the programs may invent stuff, a phenomenon now termed "hallucinations."

No comments: