Fine tuning is extra training that adapts a general AI model to a narrow job using your own examples. Few recruiting teams need it. Good prompts, a few examples in the prompt and grounding on your own data usually get you most of the way for far less cost and effort.
AItopic
2 minread
2related terms
In one line
Extra training that adapts a general model to a narrow job using your own examples. Few recruiting teams need it; good prompts and grounding on your data usually get you most of the way.
A fine tuned model starts from a foundation model and is trained further on hundreds or thousands of example inputs and ideal outputs. The result is a model that follows a specific format or style more reliably without long instructions in every prompt.
Fine tuning changes how a model behaves, not what it knows day to day. It is a poor way to teach a model your current open roles or candidate pipeline, because that information changes constantly. For facts, grounding the model in your ATS or documents is the better fit.
Fine tuning also brings work: collecting clean examples, removing personal data, testing for bias and repeating the process when the base model changes. Vendors sometimes fine tune for a specific task, such as parsing resumes, and you benefit without doing it yourself.
Why it matters to recruiters
Vendors may pitch fine tuned or custom models as a reason to pay more. Knowing what fine tuning does helps you judge whether it solves your problem. For most recruiting desks the bigger gains come from better prompts, cleaner data and solid workflows, not a custom model.
An example from a recruiting desk
A staffing firm wants AI written candidate summaries in its exact submittal format. A consultant proposes a fine tuned model. The ops lead first tries a prompt with three example summaries and the format spelled out. It gets the format right in nearly every test, so the firm saves the budget and revisits fine tuning only if volume grows a lot.
Tips that help
Try clear instructions and few shot examples before considering fine tuning.
Use grounding, not fine tuning, for facts that change such as jobs and candidates.
Strip personal data from any examples used for training.
Ask vendors who claim custom models what data they trained on and how they tested it.
Mistakes to avoid
Fine tuning to teach a model facts that will be out of date next month.
Training on past hiring decisions without checking them for bias.
Frequently asked questions
What is the difference between fine tuning and prompting?
Prompting gives instructions and examples at the moment you ask. Fine tuning changes the model through extra training so it behaves a certain way by default. Prompting is faster, cheaper and easier to change.
How much data does fine tuning need?
It depends on the task and provider, but you generally need many high quality examples of input and ideal output. Poor or inconsistent examples produce a poor model.
Can fine tuning make AI hiring decisions fair?
Not on its own. A model trained on biased past decisions can repeat that bias. Fairness needs testing of outcomes, human review and attention to which data the model uses.
