The Human Impersonator: Language, AI and GPT-3

A long time ago, I wanted to be a standup comedian and spent a lot of time hanging out, working (and yes, even performing) at Yuk Yuk's, a Toronto club. (You'll notice I shifted gears more than a few times since then ?? ??.)

My favorite acts were conceptual comics, who'd take you on a hilariously twisted roller coaster kind of ride.

I also liked observing some of the more um... observational comedians.

But when it came to impressionists, I was conflicted. On one hand, I liked the way they conjured up images of famous stars and politicians just by the sound of their voice.

On the other, I thought they had it easier than comics who wrote all their own material since an impressionist's bits were based on imitation. More reproduction than original and ultimately, kind of a fake.

Make That AI Stop Imitating Me!

That's one of the same challenges facing natural language processing (NLP), algorithms that see patterns in the way words fit together, and natural language generation (NLG), algorithms that use those patterns to predict which words should come next and churn out synthetic copy.

Right now, NLG is like the impressionist of the digital world. And it's getting pretty darn good imitating us.

Except for one major challenge.

You Want the Truth ... AI Can't Handle the Truth

With apologies to Jack Nicholson in the film, A Few Good Men, AI has a problem with truth.

You see, like any good mimic, it values the way things sound over veracity and facts.

And that becomes a major issue when we set it loose online.

If you want to see an example of the good and bad parts of NLG in action, check out the op-ed on why we shouldn't be scared of our AI overlords, penned by GPT-3, OpenAI's new model.

It's pretty convincing and it lies.

Writing the Story of GPT-3

That's because GPT-3 wasn't built to filter for truth. When OpenAI released it, they said it was trained on about a trillion words—it pretty much read the web.

And you can see the results in the impressive copy it churns out. But it can't distinguish factual from fake and has trouble with bias and harmful language.

GPT-3 is the subject of this week's Digital Marketing Trends video.

You'll hear how the model is currently being commercialized and that soon we'll find it in more marketing products and apps.

Are you ready for that?

Check out the video and let me know what you think.

February Five for Free

Before we go, I wanted to remind you about the five in-demand courses LinkedIn opened up for the month of February for users to watch for free. These include content marketing, writing copy, management skills and using Excel for finance.

And Digital Marketing Trends.

There's just over a week left to access the promotion, so if you know of anyone who might enjoy the course, but isn't a LinkedIn Learning member, please share this link with them.

Connect with Martin

And that's all the AI wrote for issue #25.

As always, thank you for reading and subscribing! I appreciate you being here and sharing your comments and thoughts. Have you had a chance to try NLG as a creative partner? What did you think of the quality of its prose? Please share your experiences in the comments below.

Be sure to reach out if you have questions about any of the videos in Digital Marketing Trends, or my other LinkedIn Learning courses.

And if you want to find me, follow or connect on LinkedIn or Twitter. (And if you are connecting on LinkedIn, please add a short note so I know how you found me.) Or visit my website and send a message or a question.

Check out the #LiLDigTrends hashtag for quick updates between videos.

And now, if you'll excuse me, I have to head out. You know ... timing.