【新智元导读】大模型的预训练-微调范式，正在悄然改写强化学习！伯克利团队提出新方法InFOM，不依赖奖励信号，也能在多个任务中实现超强迁移，还能做到「读心术」级别的推理。这到底怎么做到的？

"Accelerating the development of new quality productive forces" is a key priority in China’s 2024 Government Work Report. The integration of innovation and technology into industrial development will ...