Author name: Rachel Ma, Dylan Hadfield-Menell, Kristjan Greenewald

arXiv:2605.06785v2 Announce Type: replace
Abstract: Inference-time scaling methods rely on Process Reward Models (PRMs), which are often poorly calibrated and overestimate success probabilities. We propose, to our knowledge, the first use of condition…

arXiv:2605.06785v1 Announce Type: cross
Abstract: Inference-time scaling methods rely on Process Reward Models (PRMs), which are often poorly calibrated and overestimate success probabilities. We propose, to our knowledge, the first use of conditional…