Gemini 1.0 Pro

New RL method teaches LLMs to self-correct answers

Researchers have developed SCoRe, a novel two-stage reinforcement learning technique that enables language models to refine their own responses using self-generated data. This method significantly improves performance o…