Search papers, labs, and topics across Lattice.
1
0
3
2
LLMs struggle with long-horizon reasoning in software engineering because they retrieve irrelevant memories, but aligning memory with subtasks boosts performance by 4.7 points on SWE-bench.
