ResearchExplainer
How AI detectors work: three methods and why they often fail
Classifiers, perplexity tests and watermarks score well in the lab, but research shows rewriting cut one detector's catch rate from 70.3% to 4.6%.
Every ClaudeGPT Si story about MIT, newest first: news, analysis, reviews and guides.
1 story · Updated
Classifiers, perplexity tests and watermarks score well in the lab, but research shows rewriting cut one detector's catch rate from 70.3% to 4.6%.