The automatic mode of the Event Transform action is powered by AI. Here, we’ll guide you through how to best guide the AI to iteratively build what you need.
Characteristics of good guidance
Providing guidance to the AI is just like asking a real person for help: the clearer and simpler your request, the better the outcome is likely to be.
Explain the input. As well as your guidance, the AI will examine the input data. But you can augment its understanding of the data. For example, you might say “the input contains a base64-encoded image”, to give the AI context on what it’s dealing with.
Break down into steps. If you think the AI will need to do several things in serial to achieve your outcome, tell it. For example, you might say “First remove any blank values from the list. Then, format each value as a dollar value.”.
Be clear about the output you want. For simple outputs, use plain English (“return the severity of the alert”). For more complex data structures, type out an example of the shape of the data to be returned, e.g. with JSON.
Check your work. If you’re not achieving the results you expect, chances are your guidance is not clear enough. Read it as though you were the person answering the request, and tweak anything that feels potentially vague or confusing.
Most importantly, great guidance is obtained through iteration.
Iterating on guidance
Instead of trying to get your guidance right in one go, it’s best to start simple and work your way to the output you need.
We’ve designed our system to author and run the underlying code very quickly, so you get the feedback you need to iterate towards success.
Start small
Before you even start thinking about what output you need, consider providing generic guidance to get started, like “summarize the input”. This will give you some initial confidence that the AI is receiving and understanding the data you’re providing.
If your transformation involves several steps, perhaps start with the first step on its own
– e.g. “remove all blank values”. Or, if your transformation is a complex analysis of a list of data points, start with a simple version – e.g. “count the items in the list”.
Experiment safely
The automatic mode editor is a playground: your changes do not take effect until you press “save”. Until then, any changes or errors you introduce are safely detached from your workflow.
As you iterate on your guidance, you’ll often see errors or mistakes. Each time you spot something going wrong, think about how to experiment to improve your guidance. And don’t worry: you won’t lose your work – you can always go back and forth through your guidance history.
Read the code (if you want)
Automatic mode does not require you to be able to read or write code. But glancing at the code can give you hints on where the AI may be misunderstanding you.
Even if you’re not a programmer, go ahead and take a look – you might be surprised at your ability to help the AI, by looking at how it has modeled the task at hand.
If in doubt, rewrite
Just like when speaking to another human, sometimes the AI won’t catch your drift. Try replacing a sentence with another way of saying the same thing, and see if you can make your intent a little clearer that way.