Fine-tuning and RLHF

After pre-training, models are fine-tuned with human feedback (RLHF). Humans rank responses, teaching the model to prefer helpful, safe outputs.

• RLHF uses human rankings to improve models.
• Fine-tuning aligns models with human preferences.