Training the AI

For my entire adult life, family members have been calling me to ask their questions on anything and everything. To answer them, I usually google it, verify the information, synthesize the results, and phrase it in an easily understandable way. Imagine my luck when I was tapped to be on a team of writers tasked with taking queries, researching them, verifying the information, synthesizing the results, and phrasing the responses in easily understandable ways to train an AI to do the same. I hit the ground running.

Google was developing a generative AI to transform the way people lowercase-g google. This was a natural language processor powered by a large language model. Do I know exactly what that means? Not really. But I do know language, and that’s what AI needs. A huge amount of training data is necessary to teach an AI model to give nuanced responses to complex questions and topics, grounded in credible sources.

It needs to “learn” to use the right tone, to speak the right way, and to produce high-quality outputs. AI can’t do what it hasn’t been taught to do. And just as important, it needs to be taught what not to do. Part of a content designer’s job is to guard against style violations.

Some people have responded to this project with, “So you’re working on the thing that’ll put us all out of jobs?” Maybe, but by then I could probably use a break anyway. Either way, my family will keep calling me for the answers regardless of how incredible SGE is. But I’ll use it to answer them faster and better.