Search papers, labs, and topics across Lattice.
2
0
6
3
LLM coding agents still fall short when optimizing real-world codebases, especially when balancing multiple objectives like performance and correctness, as revealed by the new FormulaCode benchmark.
Stop hand-crafting RLHF curricula: ACTOR-CURATOR learns to dynamically select training problems, boosting performance by up to 30% and speeding up training by 80% on challenging reasoning tasks.
