Building conversational guidelines

A guideline can be built into a test case. It forces people to follow a defined solution. The testing software can test for a specific font or size in a page header or a button label that doesn’t use a term an organization forbids (such as abort, kill, or execute). Doing something precisely, the same every time isn’t very conversational. Repeatability and consistency are crucial in enterprise solutions that don’t always match a conversational style, tone, and engagement. Consider that the LLM can understand a range of phrases while a fixed traditional test suite has little flexibility, although a test could look for one phrase from a collection of options.. The testing we did in the earlier chapters is a form of validation and can be used to address if the model is acting as expected for areas like following the ten heuristics or adhering to specific guidelines.

Develop guidelines by drawing on existing resources and consider...