LLMs produce counterexamples that remove up to 11.68% of invalid assertions from dynamic inference and raise precision by up to 7% on benchmarks without hurting recall.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improving Dynamic Specification Inference with LLM-Generated Counterexamples
LLMs produce counterexamples that remove up to 11.68% of invalid assertions from dynamic inference and raise precision by up to 7% on benchmarks without hurting recall.