← sources

With regard to Hallucination Rates

source: reddit (reddit:artificial)published 2026-09-22 17:28ingested 2026-09-23 02:37

retained text

As frontier LLM models start to trend toward lower hallucination rates across all domains, I've always thought it a bit ironic that we treat hallucination rates as the major flaw that separates humans and machines. In reality, your best friend, parents, co-workers, are consistently "hallucinating" information on a daily basis. Most people are just slinging out information with high confidence and little regard to factual accuracy. The trust itself between the humans actors is evidently stronger than between man and machine, but I believe it's time to give more trust to the models' answers themselves, while always maintaining our ability to critical think and identify misinformation when it arises.

\* An overlooked caveat to this is I'm talking mainly about advanced, paid models. It should be universally known by now the free tiers are and will continue to be the weaker, dumber younger brother and when they sound incorrect, you go and ask the more competent, stronger brother for a better answer.

document fetched 2026-09-23 00:20 · canonical https://reddit.com/r/artificial/comments/1wnfyj5/with_regard_to_hallucination_rates