#Test-Time Reinforcement Learning

Researchers from Tsinghua University and Shanghai AI Lab introduce TTRL, a novel method allowing large language models to improve their performance without labeled data by leveraging self-generated pseudo-rewards during inference.