https://hackerpaper.kaipfister.de/p/2505.00506/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:27:26 GMThttps://news.ycombinator.com/item?id=43906890urn:hackerpapers:submission:2505.00506:43906890Tue, 06 May 2025 16:23:37 GMT<p>Show HN: HalluMix – A Benchmark for Real-World LLM Hallucination Detection</p><p>We built HalluMix to evaluate how well hallucination detectors perform in the kinds of messy, high-stakes environments where LLMs are actually deployed: long-form outputs, multi-document contexts, and domain-specific tasks like law, medicine, science, and news. Most existing benchmarks focus on synthetic or short-form QA data. That didn’t reflect what we were seeing in production, so we built our own to test our hallucination detectors, and decided to open source it. The dataset includes 6,500 examples across QA, summarization, and NLI tasks. We added distractor documents, shuffled the…</p><p>4 points; 0 discussion replies on HN. <a href="https://hackerpaper.kaipfister.de/p/2505.00506/">Paper and source history</a>. A submission is not an endorsement.</p>cs.CLhttps://news.ycombinator.com/item?id=43905692urn:hackerpapers:submission:2505.00506:43905692Tue, 06 May 2025 14:41:32 GMT<p>HalluMix Benchmark: Detecting Hallucinations in Real-World Scenarios</p><p></p><p>2 points; 0 discussion replies on HN. <a href="https://hackerpaper.kaipfister.de/p/2505.00506/">Paper and source history</a>. A submission is not an endorsement.</p>cs.CL