Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design.¶
Sclar, M., Choi, Y., Tsvetkov, Y., & Suhr, A. (2024). Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design. Proceedings of ICLR 2024.
Cited by¶
1 citation across 1 artifact.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Elicitation Channel Contribution
- Language-model prompting: the same model on the same query produces different outputs under different prompts; prompt engineering is elicitation-channel design, and modern evaluation randomises across templates and reports sensitivity bands.
This sourceDemonstrates that trivial prompt-formatting changes produce large output variation, motivating randomization across templates and reporting of sensitivity bands.
- Language-model prompting: the same model on the same query produces different outputs under different prompts; prompt engineering is elicitation-channel design, and modern evaluation randomises across templates and reports sensitivity bands.
Verification¶
This reference passed the adversarial substantiation pipeline: it was checked to exist and to support the claim it is attached to. See how references were verified.
Registry ID ref:16c572a46fb9 · see in the full table