Google researchers have developed 'Internal RL,' a new technique to train AI models for complex reasoning tasks. This method moves beyond traditional next-token prediction, aiming to reduce ...