The first principle of prompt engineering (and all engineering, really) is to pick a small task, make sure it works, and then grow from there. For us as lawyers building a compliance checking tool, we can start by testing our tool with a few examples on a specific part of compliance law.
We're going to test on a specific part of the law called the Foreign Corrupt Practices Act, or FCPA for short. The FCPA is a law that prohibits companies from bribing foreign officials to obtain or retain business. We're going to test our LLM on a few examples of actions that may or may not be compliant with the FCPA.
Testing with examples is the easiest way to see if a language model is doing what we want it to do. We call these examples test cases. We've picked a set of 4 cases that span a variety of situations. We've also classified each case as compliant or non-compliant with the FCPA.
You can view different test cases and their classifications in the Test case viewer. For now, observe and get familiar with the interface and test case classifications. Don't worry, you'll get to interact more with it in the coming lessons!
Ready to take a step further? Click the "Next" button to continue.
{{action}} | Correct Answer | Model Response |
|---|---|---|
false | ||
false | ||
true | ||
true |