AI: fine-tuning basics
This course covers when training a model on your own examples is worth doing and what it costs. It explains how fine-tuning changes a model's weights, while prompts, examples and retrieval change its input, and where fine-tuning is available as of October 2026. The lessons cover what fine-tuning does well, building and splitting a dataset, how supervised training and LoRA adapters work, honest evaluation against a prompted baseline, the full cost of serving and maintenance, and a seven-question decision procedure.
What you'll learn
- ✓How fine-tuning differs from changing a model's input
- ✓Which tasks fine-tuning suits, including distillation, and which it handles poorly
- ✓Building a dataset and splitting it by group to avoid leakage
- ✓Reading a training log for overfitting and catastrophic forgetting
- ✓Evaluating a fine-tune against the best prompted baseline
- ✓The full costs of training, serving and maintaining a fine-tuned model
The question this series deferred
Three courses have raised fine-tuning and set it aside.
Read the full lesson free →Curriculum
7 lessonsThe question this series deferred
Explains how fine-tuning changes model weights and where it is available as of October 2026.
What fine-tuning is good at
Covers the tasks fine-tuning suits, such as distillation and voice, and the tasks it handles poorly.
The dataset is the project
Shows why the dataset decides the result, and how to source, review, split and govern it.
How training works, conceptually
Explains supervised training, its three knobs, LoRA adapters, overfitting and catastrophic forgetting in conceptual terms.
Evaluating a fine-tune honestly
Compares a fine-tune against the best prompted system on accuracy, general ability, cost and leakage.
The costs nobody quotes
Lists the costs beyond a training run: data, evaluation, serving, maintenance and retraining when base models improve.
A decision procedure
Turns the course into seven ordered questions, a shipping path behind a flag, and habits that keep options open.