pith. the verified trust layer for science. sign in

A practical analysis of human alignment with *po, 2025

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it

fields

cs.LG 1

years

2025 1

verdicts

UNVERDICTED 1

representative citing papers

Failure Modes of Maximum Entropy RLHF

cs.LG · 2025-09-24 · unverdicted · novelty 5.0

Derives SimPO from MaxEnt RL and reports that MaxEnt RL in online RLHF exhibits frequent overoptimization and unstable KL dynamics across scales, unlike stable KL-constrained baselines.

citing papers explorer

Showing 1 of 1 citing paper.

  • Failure Modes of Maximum Entropy RLHF cs.LG · 2025-09-24 · unverdicted · none · ref 5

    Derives SimPO from MaxEnt RL and reports that MaxEnt RL in online RLHF exhibits frequent overoptimization and unstable KL dynamics across scales, unlike stable KL-constrained baselines.