Search papers, labs, and topics across Lattice.
3
0
4
2
LLMs can't rebuild software from scratch, even for widely used programs like FFmpeg and SQLite, revealing a critical gap in their ability to make high-level software architecture decisions.
Turns out, coding agents in the wild are only writing useful code 44% of the time, and are introducing more security vulnerabilities than human developers.
SLMs can leapfrog performance on complex software engineering tasks by learning *when* to ask for help from larger models, achieving a 25% gain on SWE-bench with minimal expert queries.
