In Case You Missed It: ARC 'Challenge' Is Not That Challenging
This work addresses a critical issue for AI researchers by exposing how evaluation setups can distort perceived model capabilities, though it is incremental as it builds on existing shifts in practice.
The paper tackles the problem of misleading difficulty in AI benchmarks like ARC Challenge, showing that a fairer evaluation method reduces performance gaps and yields superhuman results on tasks such as OpenBookQA.
ARC Challenge appears more difficult than ARC Easy for modern LLMs primarily due to an evaluation setup that prevents direct comparison of answer choices rather than inherent complexity. Although some researchers have quietly shifted to a more appropriate scheme over the last year, the implications of this change have yet to be widely acknowledged. We highlight this overlooked shift, show how similar evaluation practices falsely imply reasoning deficits in other benchmarks, and demonstrate that fairer methods dramatically reduce performance gaps (e.g. on SIQA) and even yield superhuman results (OpenBookQA). In doing so, we reveal how evaluation shapes perceived difficulty and offer guidelines to ensure that multiple-choice evaluations accurately reflect actual model capabilities.