Andrew Davis on what publishers don’t get about using AI
At the 2024 Niche Media Conference, keynote speaker Andrew Davis of Monumental Shift told an audience of about 400 niche publishers that reacting to AI creates an emotional roller coaster. First, it looks brilliant, then it delivers C- content, then it’s brilliant again, and then it delivers D + content. Then you wonder if the recipient will discover if you’ve used it.
Then it’s brilliant.
Then you think it’s evil and may take your job.
He said part of the problem is that we look at AI from the wrong perspective. We still compare AI to search or language simulation. Rather, he said, its superpower is the ability to mimic, well, you.
Specifically, how you, with your skill sets, would approach a very specific task. The goal should be to train AI to mimic your own unique processes, he said.
“Think of it as a digital doppelganger. It isn’t prompt engineering. It’s imparting your essence into it. Once AI grasps your own approach, it doesn’t just assist, it augments,” he said. “Think of it as a co-pilot for a very specific task. Then you are in a team helping each other.”
To demonstrate, he showed how he created his own doppelganger. He mentioned testing ElevenLabs to clone his voice, Midjourney AI to generate art from text, and Heygen for video.
But in reality, he said, it’s not about the bike. Rather, to deliver great content, he suggested that AI today has a brain roughly the size of a squirrel, which makes sense due to its seemingly anxious desire to please and the breakneck speed but only C+ answers.
To get an A+ result on a specific task, you have to feed it “nuts” – that is, the best sample of your writing, for example, so he, she, or it can learn to mimic how you, specifically, want the thing to be done.
The very specific task
To train his AI to write an introductory script for video promotion, for example, he followed various small steps in the form of a conversation. He asked the AI to pretend to be him and to write a script. Then he fed AI a couple of scripts he liked – not all at once. Then he wanted it shorter. Then he asked AI to add some humor. Each instruction got the script closer and closer to how he would have written it. Finally, he could use AI to generate a video of him talking, using the script the AI wrote.
The whole process took an hour, but it can be used repeatedly. A task that used to take up 40 hours a year in time, is now a few seconds.
While you can’t technically “save” a conversation in GPT4, as long as you keep the tab open and don’t refresh the page(update, the capability to save prompt histories is now available), you can add an additional task for your doppelganger to accomplish. If you close the tab, that work disappears.
Finally, he advised you always to disclose when you’ve used AI, so the recipient doesn’t feel fooled.
And give your doppelganger a name. This will help you treat it like the person you are training it to be: You.
