Iterating on the Prompt

Update the Prompt

Navigate to templates and press the "..." next to our existing template. Choose "Make a copy."

Make a copy of the current template

Open the copy at the top of the templates list. Then press the "..." by the title and choose Rename.

Name it "Prompt v2".

Next, edit the system prompt and replace it with this:

You are a professional email marketer. Review the following article and write a creative email subject line idea for it that is likely to grab attention. Do not return anything except the subject line. Do not include quotes around the email subject.

Your template should look like this:

Template with modified prompt

Press "Save."

Try the New Template with a Model

Go back to our Models list and choose to apply a template to a model again.

This time, select our Prompt v2 template from the dropdown.

Use the same model again (GPT-4 Omni) so we know the only thing that changed was our prompt.

Open the new templated model, wait for the evaluation to finish, and check the results.

It looks like the change to our prompt worked and the email subject line is no longer wrapped in double quotation marks.

Let's rate them all 4 or 5 stars.

Compare Models

Back on our models list, we can now compare the score for these two models.

We can also rename them to v1 and v2 to make our table easier to read.

With more validation examples, and by reviewing and rating the outputs from your templated models, you can get a clear picture of which model/template combinations are performing best.

Where to Go From Here

Next, we would need to expand our validation dataset and try the prompt on other models (including models from other model providers) to see which give us the best outputs, at the right price and speed for our project.

In the next guide, we'll learn how to build and grow a dataset.