What is the difference between alignment and fine-tuning in LLM training?

Fine-tuning adapts model weights to tasks, while alignment shapes behavior to match human preferences.

Fine-tuning optimizes task-specific loss (e.g., classification or instruction following). Alignment focuses on behavior shaping using RLHF, DPO, or constitutional methods. Alignment ensures safety, helpfulness, and adherence to human intent rather than raw task performance.