Cloze testing is a common method for measuring the behavior of large language models on a number of benchmark tasks. Using the MMLU dataset, we show that the base-rate probability (BRP) differences across answer tokens are significant and affect task performance ie. guess A if uncertain. We find that counterfactual prompting does sufficiently mitigate the BRP effect. The BRP effect is found to have a similar effect to test taking strategies employed by humans leading to the conflation of task performance and test-taking ability. We propose the Nvr-X-MMLU task, a variation of MMLU, which helps to disambiguate test-taking ability from task performance and reports the latter.
Cite
@article{arxiv.2406.11634,
title = {The Base-Rate Effect on LLM Benchmark Performance: Disambiguating Test-Taking Strategies from Benchmark Performance},
author = {Kyle Moore and Jesse Roberts and Thao Pham and Oseremhen Ewaleifoh and Doug Fisher},
journal= {arXiv preprint arXiv:2406.11634},
year = {2024}
}