“While technically true the hallucination rates on modern models is low…”

Isn’t this entirely context dependent? Where did you get the information that modern models have low hallucination rates? I’d love to see the benchmark if there is one, it seems like it would be useful to track.