The biggest problem with AI models right now is not that it’s “dumb” – it’s that it’s a confident liar
Most models would rather confidently lie than admit they don’t know something
Grok is actually solving this..
The Hallucination Rate measures incorrect answers against total https://t.co/ED6o1S3uc5
— X Freeze (@XFreeze)
Apr 25, 2026
from http://twitter.com/XFreeze
via IFTTT