Is Your Large Language Model Knowledgeable or a Choices-Only Cheater?
This addresses concerns about the validity of MCQA leaderboards for evaluating LLMs, showing that high performance is not merely due to exploiting choices-only shortcuts, though it is incremental as it builds on prior work on contrast sets.
The study investigated whether large language models (LLMs) rely on choices-only shortcuts to achieve high rankings on multiple-choice question answering (MCQA) leaderboards, finding that 12 tested LLMs did not exhibit such reliance when given both questions and choices, with results validated on an 820-question contrast set.
Recent work shows that large language models (LLMs) can answer multiple-choice questions using only the choices, but does this mean that MCQA leaderboard rankings of LLMs are largely influenced by abilities in choices-only settings? To answer this, we use a contrast set that probes if LLMs over-rely on choices-only shortcuts in MCQA. While previous works build contrast sets via expensive human annotations or model-generated data which can be biased, we employ graph mining to extract contrast sets from existing MCQA datasets. We use our method on UnifiedQA, a group of six commonsense reasoning datasets with high choices-only accuracy, to build an 820-question contrast set. After validating our contrast set, we test 12 LLMs, finding that these models do not exhibit reliance on choice-only shortcuts when given both the question and choices. Thus, despite the susceptibility~of MCQA to high choices-only accuracy, we argue that LLMs are not obtaining high ranks on MCQA leaderboards just due to their ability to exploit choices-only shortcuts.