As part of the AI Academy for Marketers membership, we offer monthly Ask Me Anything (AMA) coaching sessions with leading industry experts. We chat about everything from technology trends, use cases, lessons learned, to much more.
This post offers an inside look into the latest, exclusive members-only session with Jeff Coyle (@jeffrey_coyle), Co-Founder and Chief Strategy Officer, MarketMuse. During the session, Jeff discussed current and future trends surrounding natural language generation—all questions stemming from his courses offered in AI Academy for Marketers (linked below).
Below is a quick video from our chat, followed by top takeaways from the conversation.
Basically, natural language generation (NLG) takes structured data and turns it into a language (i.e. English). It's not defined specifically; it is the building of a language model. The language model generates text based on the model. That is what GPT-3 is too—a specialized language model with a goal to produce text. GPT-3 is created by OpenAI. Our model is MarketMuse First Draft. GPT-3 is very generalized with no specific use case tied to it. It's a language model that allows you to think of a use case and see if it works. And yes, It IS hype-worthy; it's an amazing feat of engineering!
To work with NLG, or to get your own prompt, it does not take much at all. You just need to make a great business case for it. NLG is in the early days, though, so it can change. But getting a subscription to a tool, like to MarketMuse First Draft, requires no expertise! You can simply make an outline and get a draft. Some are even simpler than that.
The MarketMuse Content Brief allows you to check your work to see if you have any blind spots. If you are consistently writing the best article, call me. Otherwise, building the brief provides additional value in various ways—like recommendations for titles, subheads, etc. For example, if the article is about NLG, you can add a variant like “NLG solution.” The brief will give you prompts to include in your draft, like internal and external linking recommendations. Sometimes you can use it to find your blind spots and enrich your existing content with more depth of information. It's not a replacement for the human; rather, it can be a starting point or to add value to an existing piece.
It is tunable for style, tone, usage and structure. It is very expensive though.
By default, our system (MarketMuse) works to try to know the most in the world about the concepts you’re searching for. You can, when building the model and when tuning it, use specific sources. You can also train the model to write like Paul (for example) or only use biased content.
With NLG, what we are going to see is (I hope) collaboration among the major innovators in the space to provide protection for the market and make extreme advancements. I would also predict that the template-based solutions are going to realize they need to get going with these advanced solutions and do a better job. I think Google will be at the forefront; their teams are brilliant. Apple and Salesforce will definitely be in the mix too. Overall, recommendation engine products, personalization products and really useful generation products will probably exist by the end of this year. MarketMuse will be one of the best ones!
Learn more about the topics discussed in this post with a membership to AI Academy for Marketers. The Academy is our members-only online education platform and community that features dozens of on-demand courses and certifications taught by leading AI and marketing experts.
Click below to learn more about AI Academy for Marketers below.