Hallucination rates are vanity. A model may pass one test but hit 30% errors on...

https://www.mediafire.com/file/ex8a71ev923hbwn/pdf-22694-13739.pdf/file

Hallucination rates are vanity. A model may pass one test but hit 30% errors on HalluHard. Because Vectara’s HHEM and AA-Omniscience measure facts differently, your yardstick is key