LLMs generate compilable but semantically weak tests for unseen proprietary systems like SAP HANA while performing better on open-source LevelDB, indicating reliance on shortcuts rather than robust reasoning.
Martha Lewis and Melanie Mitchell
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
-
LLMs taking shortcuts in test generation: A study with SAP HANA and LevelDB
LLMs generate compilable but semantically weak tests for unseen proprietary systems like SAP HANA while performing better on open-source LevelDB, indicating reliance on shortcuts rather than robust reasoning.