AutoCBI uses an LLM to merge four sources of information about a failing compilation, and re-ranks suspicious compiler files, isolating the true buggy file at rank 1 for 42 of 120 real GCC/LLVM bugs.
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
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Compiler Bug Isolation by Leveraging Large Language Models
AutoCBI uses an LLM to merge four sources of information about a failing compilation, and re-ranks suspicious compiler files, isolating the true buggy file at rank 1 for 42 of 120 real GCC/LLVM bugs.