A benchmark for offline RL with LLM-generated imaginary rollouts shows existing algorithms reach only 35.44% success on hard tasks versus 64.37% when trained on real rollouts.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ImagineBench: Evaluating Reinforcement Learning with Large Language Model Rollouts
A benchmark for offline RL with LLM-generated imaginary rollouts shows existing algorithms reach only 35.44% success on hard tasks versus 64.37% when trained on real rollouts.