Search papers, labs, and topics across Lattice.
3
0
7
13
Strong logical planning in LLM agents doesn't ensure safe execution鈥擯eakBench reveals critical flaws in existing benchmarks that could lead to resource overflows.
LLMs still struggle to go beyond simple lookups when answering questions about tables, especially when prediction and reasoning about unobserved data is required.
Forget hand-tuning rollout budgets: $V_{0.5}$ dynamically allocates compute to sparse RL rollouts based on a real-time statistical test of a generalist value model's prior, slashing variance and boosting performance.
