New hallucination benchmark accepted to NeurIPS 2026 Evaluations & Datasets track (X thread by @_emliu)

Summary

An X thread from @_emliu announcing a new hallucination benchmark accepted to the NeurIPS 2026 Evaluations & Datasets track, built around a 'controlled world' where facts can be verified. The benchmark's name, paper, and code were not independently verifiable from the thread alone.

Keywords

hallucination · benchmarks · evaluation · neurips · x-thread

Topics

hallucination, benchmarks, evaluation, neurips, x-thread

Research notes

• Discovery: Posted in #random-papers on 2026-09-28. Thread text partially visible in the Discord embed: 'Excited to share our new benchmark is accepted to NeurIPS 2026 Evaluations & Datasets track! ... controlled world ...' — about a hallucination benchmark.
• Limitations: Thread text truncated in the Discord embed; benchmark name, paper, dataset, and results were not independently verified via search. Acceptance claim is unverifiable from available sources (NeurIPS 2026 proceedings not yet public at check time).
• If the benchmark paper surfaces later, upgrade this to a paper entry with canonical URL; keep the thread as the discovery source.
• STATUS=ambiguous: verify before relying on this entry.