开源大模型新王干翻GPT-4o，新技术可纠正自己幻觉，数学99.2分刷爆测试集

Reflection 70B: A Ground Breaking Open-Source LLM, Trained with a New Technique called Reflection-Tuning that Teaches a LLM to Detect Mistakes in Its Reasoning and Correct Course