Binoculars
A zero-shot AI-generated text detector that identifies text provenance by comparing metrics such as model log-likelihoods. It is relatively sensitive to variation in text length.
Timeline (last 90 days)
Taking the Binoculars detector as an example, a single threshold set at a length of 200 tokens causes the false positive rate to soar to 29% when the text is shortened to 10 tokens.
On three zero-shot detectors, three corpora and two open-weight generators, averaged over 300 random document-level repartitions, binning calibration stabilized the average false positive rate at 5.0% in every detector-length combination from 10 tokens to 200 tokens.
In the experiments, no detector reached an 80% true positive rate; the best-performing one, Binoculars, reached only 59% at 200 tokens.
All Facts (3)
Taking the Binoculars detector as an example, a single threshold set at a length of 200 tokens causes the false positive rate to soar to 29% when the text is shortened to 10 tokens.
50%UnverifiedOn three zero-shot detectors, three corpora and two open-weight generators, averaged over 300 random document-level repartitions, binning calibration stabilized the average false positive rate at 5.0% in every detector-length combination from 10 tokens to 200 tokens.
50%UnverifiedIn the experiments, no detector reached an 80% true positive rate; the best-performing one, Binoculars, reached only 59% at 200 tokens.
50%