The technique is called reinforcement learning from human feedback (RLHF). It's how, for example, Facebook can train Llama models to meet a certain safety score

The technique is called reinforcement learning from human feedback (RLHF). It's how, for example, Facebook can train Llama models to meet a certain safety score before releasing them into the wild.