Sclar et al. 2024

Quantifying language models' sensitivity to spurious features in prompt design. Sclar, M., Choi, Y., Tsvetkov, Y., and Suhr, A. (2024). ICLR.

Relationship to this project — Prompt sensitivity

Establishes that prompt formatting matters broadly. X1/X2 here differ by prospective registration, a formally game-equivalent strategic task, and a held-out minimal-span confirmation of one atomic wording-and-position operation.