Fine-Tuning
Simple Definition
Fine-tuning takes an existing pre-trained AI model and trains it further on a smaller, specialized dataset. The result is a model that keeps all the general knowledge from pre-training but gains specialized skills or a specific style.
Think of it like taking a generalist professional and sending them to a specialized training program — they don’t forget everything they knew, they just get better at one specific thing.
Why Fine-Tune Instead of Just Prompting?
Prompting (writing detailed instructions) can adapt a model’s behavior without any training. Fine-tuning goes deeper:
| Prompting | Fine-Tuning |
|---|---|
| No training needed | Requires training data and compute |
| Works for most tasks | Better for consistent style/tone |
| Instructions in every request | Style is baked into the model |
| Easy to change | Requires retraining to change |
For most use cases, prompting is enough. Fine-tuning is valuable when you need consistent output style at scale, or when the model needs specialized domain knowledge.
Common Use Cases
- Customer support bots — trained on a company’s specific products and policies
- Medical AI — trained on clinical notes to use correct terminology
- Code assistants — trained on a specific codebase or language
- Brand voice — trained to write in a consistent tone for a company
Fine-Tuning vs. RAG
Both fine-tuning and RAG (Retrieval-Augmented Generation) add specialized knowledge to a model. Fine-tuning bakes knowledge into the model’s weights. RAG retrieves relevant documents at query time and passes them as context.
RAG is more flexible and easier to update. Fine-tuning is better for style and tone consistency.
Related Terms
- Foundation Model — the starting point for fine-tuning
- Training Data — the dataset used for fine-tuning
- RAG — an alternative approach to specializing AI with external data
- LLM — the type of model most commonly fine-tuned
See AI terms in action
Browse practical AI workflows that use the concepts in this glossary.
Last updated: