Speaker (Jeff For Local Marketing) argues that AI failures often stem from poor data quality and insufficient data management, not the AI itself. He recommends collecting or creating relevant content (text, video, audio) from an "ideal customer" or the creator, using that to train models so outputs match the desired voice/phraseology, and then evaluating outputs (e.g., blog posts). He uses the metaphor of a new employee given keys but not trained to illustrate the point.
Chapters
The next one is data issues. This is poor data quality. Maybe, like, it's insufficient and irrelevant data, or, excuse me, insufficient relevant data and lack of proper data management are often significant barriers to training effective AI models. So as an example, and I share this with my clients when we're using AI and they say, well, I remember early on, they say, well, yeah, sounds good, but it just doesn't sound like me. So what we would do is we would start getting content or creating content from that ideal customer. This video, it's written, whatever it is, audio. And we would train our models to say, we want this to sound just like Jeff. And using the same phraseology, the same terms, the same tone. And once we put that in there and we trained it to do that, and then we ask it to perform a task like writing a blog. And then I would read that blog, and I'm like, oh, my gosh, that is so good. Well, you can do the same thing and have those same oh, my gosh moments. This is so good. But not if you don't train it. It's almost like having a new employee and you give them keys to the castle and you walk away, and they don't really know what you want to do. They're not acting like you.
