What is Fine-Tuning?

Fine-Tuning is the process of training a pre-trained model further on a specific dataset to change behavior for a task or domain. Fine-Tuning can improve performance for targeted outputs but can also introduce new failure modes.

Quick definition

Fine-Tuning means training an existing AI model on additional data to specialize the AI model.

How Fine-Tuning works

• Fine-Tuning adjusts model parameters using supervised or preference-based objectives.
• Fine-Tuning can change tone, format, and task performance.
• Fine-Tuning can affect factuality and citation behavior indirectly.
• Fine-Tuning can contribute to model drift across versions.

Why Fine-Tuning matters

Fine-Tuning matters because fine-tuning changes what users can expect from model outputs.

Fine-Tuning also affects monitoring because a fine-tuned model can behave differently from a base model.

Example use cases

• Training a model to follow a specific output format.
• Specializing a model for a narrow domain vocabulary.
• Evaluating whether fine-tuning increases consistency for a prompt set.

Related terms

• Large Language Models (LLMs)
• Model Drift
• Prompt Engineering
• System Prompt
• LLM Answers