Search papers, labs, and topics across Lattice.
University of Illinois Urbana鈥揅hampaign
4
0
4
A robust multi-agent scaffold can unlock latent capabilities in fixed models, enabling a remarkable 67.4% issue resolution rate on SWE-bench Pro鈥攐utpacing the previous best by over 8 percentage points.
Turns out, LLM agents are surprisingly bad at following plans, often preferring their own internalized (and potentially flawed) workflows, and a bad plan is worse than no plan at all.
Forget wrestling with language-specific tooling: ReCodeAgent autonomously translates and validates entire code repositories across diverse languages with a 60% boost in test pass rates.
LLMs can generate surprisingly effective C unit tests when guided by program structure and constraints, achieving coverage comparable to symbolic execution while producing more readable code.
