35 AI models tested across 172 billion tokens of real document questions.
median hallucination rate:
25%.
One out of every 4 answers contain false information.
https://image.nostr.build/47fe813cdf09...
But we out here using it for target selection in WW3.