Search papers, labs, and topics across Lattice.
1
0
3
Metropolis-Hastings sampling for LLM policy composition consistently outperforms SIR, offering a provably better approximation of desired reward trade-offs with the same computational budget.
