Cleaning LLM Truthfulness Benchmarks
This matters to anyone evaluating LLMs because it shows a common benchmark may be flawed, potentially leading to overestimation of a model's reasoning capabilities.
This matters to anyone evaluating LLMs because it shows a common benchmark may be flawed, potentially leading to overestimation of a model's reasoning capabilities.