From the course: AI-Powered Development with the Anthropic API

Unlock this course with a free trial

Join today to access over 24,800 courses taught by industry experts.

Evaluating prompts with the Workbench

Evaluating prompts with the Workbench - Claude Tutorial

From the course: AI-Powered Development with the Anthropic API

Evaluating prompts with the Workbench

- Once you have your prompt working just right, it's a good idea to test and evaluate your responses at volume. For that, we can use the evaluate tab, which will give us a way of running multiple prompts in order to test our application. Here you can show the original prompt as well as show or hide ideal outputs. Notice that you can also control variables right here and modify them for each of the different runs. I'll go ahead and hide ideal outputs for now, and we can add rows manually by hitting the add row button and then type in any of our variables that we need. So I'll do three days right here, and if we want to, we can run this prompt to see how the model outputs. Now we got a different response and we can do that as often as possible. We can also generate a test case and the tool will automatically fill out our variables for us. Now I have only one variable, so that's not particularly hard, but if you include more than one variable, it will add all the variables you need at…

Contents