A rubric-graded code comprehension benchmark shows within-task accuracy gains after fine-tuning on data derived from the same benchmark, confounding the claim of true semantic understanding.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
A Code Comprehension Benchmark for Large Language Models for Code
A rubric-graded code comprehension benchmark shows within-task accuracy gains after fine-tuning on data derived from the same benchmark, confounding the claim of true semantic understanding.