Search papers, labs, and topics across Lattice.
The paper identifies a failure mode in LLMs called "prompt-variant output-mode collapse," where semantically equivalent prompt paraphrases cause models to deviate from the requested output format (e.g., bare labels). They introduce PARACONSIST, a benchmark with 900 prompts, to quantify this phenomenon by measuring answer consistency, semantic similarity, and length stability across prompt variants. Experiments on several LLMs show that only ~22% of closed-form variant responses preserve the ground-truth label, highlighting a significant vulnerability to paraphrasing.
LLMs can be surprisingly brittle: simply rephrasing a prompt, even while preserving its meaning, can cause them to completely abandon the requested output format.
When the substantive content of a request is rewritten, do large language models still answer in the format the original task asked for? We find that they often do not, even at temperature zero. On a 150-query evaluation over five compact 2025-era LLMs and four task types, we observe a systematic failure mode we call prompt-variant output-mode collapse: when a closed-form prompt asks for a bare label or a single choice token, content-preserving prompt variants can push the model into conversational prose, the requested format dissolves, and exact-match evaluation pipelines silently misjudge the result. To make this measurable, we release PARACONSIST, a 900-prompt benchmark of 150 base queries with five lexical, syntactic, and semantic-expansion prompt variants each, and a Semantic Consistency Score that decomposes prompt-variant robustness into answer consistency, sentence-BERT semantic similarity, and length stability. Under a whole-word answer-set match, only ~22% of closed-form variant responses preserve the ground-truth label inside their output, while ~78% drift away from the answer space entirely. In our pool, the dominant predictor of collapse is task structure rather than model identity, with model differentiation jointly carried by answer consistency and length stability. Robustness audits should therefore track response-mode preservation as a first-class reliability target alongside answer accuracy.
