Skip to the content.
Reference This entry is primarily explanatory reference: what the term means, why it exists, and how it is used.

Fine-tuning


In one sentence

Fine-tuning continues the training of a pre-trained model on task-specific examples, updating all or some of its parameters so that it performs a desired task or behaviour more reliably.

Why fine-tuning exists

A pre-trained language model is a generalist. It may not follow a firm’s conventions, produce a required output reliably, or handle a specialised task well. Three common interventions are:

  1. Prompt engineering — supply clearer instructions and examples at query time.
  2. RAG — retrieve relevant material at query time and place it in the model’s context.
  3. Fine-tuning — train on examples so the model’s parameters change.

Fine-tuning is the heaviest of the three. Most projects should start with the lighter options and only reach for fine-tuning when those genuinely run out of road.

What it actually does — concretely

Pre-training a frontier model from scratch requires enormous datasets and compute. Fine-tuning is the cheaper sibling: take an already-trained model and continue training it for a smaller number of steps on a focused dataset.

Three common varieties:

Where fine-tuning genuinely beats RAG

RAG is brilliant for “the model needs to know things from my corpus.” Fine-tuning is the right move when one of these is true:

If none of these apply, prompting, retrieval, application logic, or structured-output controls may solve the problem with less machinery.

Working example — a hypothetical for an Isenberg context

Imagine a management department wanted an AI tutor for case-method discussion with a defined level of rigour and a consistent Socratic style. The pieces might be:

The pre-conditions are non-trivial: someone has to define good tutoring, curate examples, obtain the necessary permissions, protect student records, and design an evaluation capable of proving improvement. The data and evaluation work, again, are the hard parts.

Why this matters in a teaching context

For BBA and MBA students, the strategic point is that model adaptation has become more accessible through vendor services and open-source tooling. The managerial difficulty has therefore moved towards deciding whether to fine-tune, securing suitable data, and evaluating the result. Product availability changes quickly; it should not be treated as part of the definition.

The strategic point worth surfacing in class: the bottleneck has moved from compute to data quality and evaluation. Most organizations that fail at fine-tuning fail because they had garbage training data or no honest way to measure whether the fine-tuned model actually does better than the base model on real tasks. Both problems are organizational, not technical.

Fine-tuning vs. RAG — when to use which

  Fine-tuning RAG
Speed to change Requires another training run Update or re-index the corpus
Current information Limited to training and other model context Can retrieve the current corpus
Main data risk Training examples may be memorised or exposed through the training pipeline Retrieved material is exposed to every component serving the query
Best for Style, tone, domain language, output formats Private knowledge bases, current data
Model size impact Can let you use a smaller, faster model Generally needs a capable base model

The two are not mutually exclusive. Many production systems use both: fine-tune the model for tone and behaviour; layer RAG on top for current knowledge. They solve different problems.

Trade-offs

Sources


RAG · Embedding

Return to Dictionary All Entries (A–Z) For Students Other Writing Capstone 2.0