According to a team of computer scientists, it doesn't take a team of computer scientists to fool many of the most popular AI-detection tools. New research from the University of Pennsylvania shows that many AI detectors are easily fooled with simple tricks and that many open-source models for detecting AI content use “dangerously high” default false positive rates. One of the study’s authors, Chris Callison-Burch, told EdScoop that one upshot of his research is that professors should think twice before accusing students of unethical behavior.
No comments:
Post a Comment