BooBench: A Halloween Benchmark That Scares LLMs and OCR Tools Into Character-Repeating Meltdowns

xennygrimmato_ · x · 2026-11-01

Jerry Liu's team released Boobench, a fun Halloween benchmark designed to "scare" LLMs and OCR tools into repeating characters endlessly — mirroring a real-world failure mode where repeated characters in documents send models completely off the rails. The dataset and eval are open-sourced on GitHub and Hugging Face.

Original post →

More from Fun

Fun channel →