Based on the score of your task, you may need to adjust some details to make it more accurate. You can optimize your task by adding more instructions to the AI about the task.

  1. Select the magnifying glass icon on the sidebar of VSCode.
  2. In the search bar type class [task name].
  3. You’ll likely end up with 3 search results. Find the result that is your task name exactly (it will not have input or output at the end).
  4. Locate the instructions variable.
  5. Type any instructions you want to add between the set of “”
  6. Rescore the task.
  7. Repeat as needed to get your desired accuracy score.

At this time, you can also experiment with adjusting models and temperatures to see that affect on scoring. To adjust one of these from the default add -r [adjustment] at the end of workflowai task playground [task name].

Ex: -r "model=gpt-3.5-turbo-0613" or -r "temperature=0.5". Multiple flags can be used together.

Example scenario (cont’d): You want to make it clear to the AI that the arrival time for flight events should be much earlier. In order to note this you could add instructions like: assume the default arrival time for flight events is 2 hours.

Stuck on anything? You can always type workflowai --help or workflowai task --help to get a list of available commands.