Fine-tuning is the process of taking a pre-trained foundation model and continuing to train it on a smaller, more specific dataset to adapt its behavior for a particular domain, task, or use case. Rather than training a model from scratch — which requires enormous resources — fine-tuning starts from a model that already understands language and adjusts its weights further based on new examples.
The result is a model that retains its general capabilities but behaves differently in the areas the fine-tuning targeted. A general-purpose language model fine-tuned on medical literature, for instance, becomes more fluent and accurate in medical contexts without losing its broader language abilities.
Fine-tuning changes the model itself. This distinguishes it from approaches like RAG or prompt engineering, which shape model behavior at the moment of a request without touching the underlying weights.
Why This Matters for Technical Writers
Fine-tuning is relevant when documenting AI products that have been customized beyond the base model’s default behavior. If a product has been fine-tuned, its outputs reflect both the foundation model’s training and the additional training layer on top, affecting what it knows, how it responds, and where its limitations lie. Understanding fine-tuning helps technical writers ask the right questions about what was changed, why, and what that means for the behaviors they are responsible for documenting accurately.
Common Confusion
Fine-tuning is frequently confused with RAG, but they solve different problems in different ways. RAG retrieves external information at the moment of a request to ground the model’s response. Fine-tuning adjusts the model’s weights before deployment to change how it behaves by default. RAG keeps the model’s knowledge current and auditable; fine-tuning shapes the model’s tendencies and tone. A product can use one, both, or neither depending on its requirements.
Related Terms
Foundation model, large language model (LLM), weights, RAG, prompt engineering, inference, training, grounding