Search papers, labs, and topics across Lattice.
4
0
5
4
Retrofitting clone detection with semantic awareness slashes irrelevant clones by up to 99% when porting code across redesigned software projects.
LLMs struggle to generate correct COBOL code, but a simple compilation-guided repair loop can boost success rates by over 50%.
GPT-4o can't even compile COBOL, but COBOL-Coder, a new domain-adapted LLM, achieves a 74% compilation success rate, finally bringing modern AI to legacy mainframe systems.
Seemingly impressive results on benchmarks like SWE-bench may be inflated by weak test suites that accept incorrect patches, with a new analysis showing top repair agents' performance dropping by up to 9% when evaluated against more robust tests.
