AA-Omniscience Hallucination Rate (lower is better) measures how often the model answers incorrectly when it should have refused or admitted to not knowing the answer. It is defined as the proportion of incorrect answers out of all non-correct responses, i.e. incorrect / (incorrect + partial answers + not attempted).
Grok 4.2 which was just released in the API just benched the best at this benchmark.
I don't see the problem with this. The chatbot is the most important part of Grok, so it makes sense Elon would be dogfooding it then providing suggestions.. He wants it to be truthful... It was shown on benchmarks recently that it hallucinates the least...
I've found cannabis to be extremely helpful. It adds a tinge of paranoia - so if you're paranoid about not reaching your potential, it can kick you into gear.