Everyone is talking about reliability, but 2026 benchmarks reveal a real mess....

https://sticky-wiki.win/index.php/The_Illusion_of_Provenance:_Dissecting_the_Citation_Failures_of_Perplexity_and_Copilot

Everyone is talking about reliability, but 2026 benchmarks reveal a real mess. Error rates swing wildly depending on the test. Even with web search, the HalluHard benchmark shows a 30.2% hallucination rate