Meet Trey Griffin, Senior Manager: AI Industry Analysis

Tell us about your role. How often do you use AI in your day-to-day?

I think I’ve got the most interesting job in the company. My job is to monitor the AI industry, identify opportunities for creators, understand what the threat environment looks like around AI, and advise leadership on trends and developments that are of concern to our creators. I read a lot of news articles and papers, trying to read the tea leaves and see where things are going.

I take recordings of webinars that I attend and use a speech-to-text AI. I use another AI to summarize and create notes for me. I still take handwritten notes too because I don’t trust the robot 100% yet. But there are so many opportunities. 

How did you become interested in AI? 

When I was in grad school for political science, I read the Transformer paper from Google while looking for something new and different for a master’s thesis. My idea was that this was going to be like the next space race. I remember having a conversation with one of my graduate advisors and he said “that seems a little too sci-fi to me.”

From there, I went into market research and got on the GPT-3 beta in 2021. I showed it to the company where I worked and said it’s not today, but pretty soon, you’ll be able to write your discussion guides and your surveys with this kind of tech. And they looked at me like I had an alien on my shoulder. I saw an opportunity to make a startup that could use this technology, and that’s what I did prior to joining Raptive.

What are you excited about when it comes to the future of AI? 

Right now you have a large language model that sits on ChatGPT, and it doesn’t do anything until you say, hey, can you help me with an outline? Or, hey, can you summarize this for me?

We’re moving towards the idea of taking these large language models and having them do things in the world as individual agents: retrieve weather information, make changes to a web page, or fill your Instacart for you. It’ll be able to go out onto the internet or the real world and identify trends and patterns and bring those back so you can take action on them.

Eventually, you’ll be able to say hey, I need to identify what trends in this creator niche are dominating the search rankings so that I can write about them, and you’ll have an agent that goes out and brings that to you every morning. That’s probably going to be what we see in the next two years.

What are some of the most common misconceptions you see about AI?

The joke about large language models is that they’re just ‘spicy autocomplete.’ You give it a prompt and it’s able to very, very accurately predict what the next word should be. And that’s not all that different from when your phone can guess what the next word is. 

But when the AI starts to have a really strong contextual understanding of all the words that came before it, all the words it’s written, and what the next word should be, you’re starting to see something that’s a lot more similar to how you sit down and write. 

It’s not understanding in the way that you or I would understand, but it can guess what the next token is. And this is way more than just completing the next word in the sentence. If you could guess the next word that’s going to come out of my mouth with 95% accuracy, that would be a superpower, right?

These models are able to really impersonate someone’s style. If you ask it to write something in the style of Stephen King, it could do that really convincingly. So it’s able to predict the next token that you write, building a model of your writing style and how you write. It’s beyond copying you—it’s almost impersonating you. That’s way more than spicy autocomplete. 

People automatically associate a high technical barrier with getting value out of AI, and that’s not the case.

Large language models are actually the simplest of the recent technologies to figure out. If you can give instructions to a five-year-old, then you can use a large language model. I really encourage people to get in there and play with it. You’ll learn very quickly what its strengths and weaknesses are.

How is Raptive using AI to benefit creators?

There are a lot of opportunities to simplify the way creators get their job done. Of course, our creators know and love Topic, but we have different products in the pipeline right now to help them turn ten-minute tasks into two-second tasks.

We do a ton behind the scenes with machine learning to optimize ads and the tech stack so things work better and everybody makes more money. That doesn’t get as much hype because it’s not as exciting to say we’re using machine learning to optimize your ad revenue, but we’ve been doing it far before Chat GPT was a twinkle in anybody’s eye. 

The products we come out with over the next year and our existing products are all going to improve because making things with this technology is getting easier and easier.

What can creators do now to protect themselves from any possible negative impacts of AI? 

Building relationships with your audience is probably the most important thing. A lot of folks rely on Google search to drive traffic to their website, and that could change in a big way. Diversify how you’re getting people onto your website, get on social media more, and find ways to connect with your audience through newsletters. Do things like Q and As, live events, premium video content for subscribers—get people to come back and see what you are doing next, not just for the next recipe.

You should also be using the tech. While it’s not coming to replace you, the person in your field competing with you, who is using it while you refuse to, is coming to replace you. I wouldn’t be worried about someone getting brand partnerships because they’re pumping out AI-generated content. But I would be worried about somebody who is getting brand partnerships because they’re using AI and creating more interesting and better content because of it.