What is SFT?
SFT stands for Supervised Fine-Tuning.
Training a model on example prompts paired with ideal answers written by people. Writing those answers is a common task for domain experts.
Supervised fine-tuning teaches a model by example. People write prompts and the ideal answer to each, and the model is trained to reproduce that quality. It usually comes before RLHF: SFT shows the model what good looks like, RLHF then sharpens its judgment between options.
What the work looks like
SFT tasks ask you to write. You might get a prompt such as "Explain the difference between a lien and a mortgage to a first-time buyer" and write the answer a senior professional would give, at the length and tone the guidelines set. Some projects ask you to write the prompts as well, aimed at the gaps in what the model knows.
What reviewers check
Accuracy first, then whether the answer follows the style guide: structure, length, no filler, and citations where required. A correct answer that ignores the format is usually sent back.
Example
A chemistry SFT task asks for a worked solution to a buffer pH problem. The golden answer shows each step, states the assumption that activity equals concentration, and gives the final pH to two decimal places.
More terms: how models are trained
Full glossaryReady to look at roles?
Every listing shows its pay and who it is open to.