Amplifying your team’s potential with generative AI

(Spoiler: It all comes down to knowing your prompts)

NLP (Natural Language Processing) models are the backbone of groundbreaking technologies like ChatGPT, DALL-E and Midjourney.

But on the user-end, they rely on text inputs to create more efficient, effective and imaginative content.

These inputs are prompts (strings for your computer programmers out there) that are usually 400 characters long and crucial to Generative AI applications and their outputs including:

• Chatbot Q and As
• Text for articles
• SEO recommendations
• Images
• Music
• Videos
• Advice
• Books

But how does it all work?

Let’s ask ChatGPT directly.

NLP Models: How Do They Work?

Prompt: Explain NLP models to an audience of digital executives for an article. Keep it under 100 words.

Chat GPT response: Natural Language Processing (NLP) models are a type of artificial intelligence (AI) that enables computers to process and understand human language. NLP models have been around for several decades, but recent advancements in deep learning and neural networks have made them more powerful than ever before. Generative AI, on the other hand, is a type of AI that can create original content, such as text, images, and music, without human intervention.

Simply put, what you put in the applications generates what you get out — with the only limit being your imagination.

Prompt Design 101

In the past, APIs (Application Programming Interfaces) relied on code from developers to structure their outputs.

With NLP models, that code is replaced by written text, making the technology attainable to teams who may not have known how to use it before.

Computers translate and interpret the text to produce their results and the prompt inputs are weighted.

The order you put things in, and how the words are structured and bundled together, matter as the models scan the prompt for clues about what the user wants to see as the output.

Some examples:

Prompt: "Write a short story about a person who discovers they have the power to time travel."

Response: "Sarah had always felt like something was missing from her life. She didn't know what it was until one day, she discovered she had the power to time travel. With this newfound ability, Sarah set out to explore different eras and find the missing piece of her puzzle. Along the way, she learned valuable lessons about life and love that she would never forget."

Prompt: "Write a poem about the beauty of nature in the fall season."

Response: "The leaves change color, a sight to behold,

As nature prepares for the winter's cold.

The air is crisp, the sky so blue,

Fall is the season that makes all things new.

The beauty of the season, a sight to see,

As nature displays her artistry."

Prompt: "Write a conversation between two people discussing the impact of technology on society."

Response: "Person A: Do you ever wonder about the impact of technology on society?

Person B: Yes, it's something that's been on my mind lately.

Person A: I mean, it's brought us so many conveniences and made our lives easier, but at what cost?

Person B: I think it's important to consider the potential downsides, like the way technology can be addictive and the impact on privacy.

Person A: And what about job loss due to automation?

Person B: That's definitely a concern. But at the same time, technology has also created new job opportunities in fields like artificial intelligence and robotics.

Person A: It's a complex issue, for sure. We need to find a balance between embracing technology and being aware of its potential consequences."

As you experiment with NLP models, you will start to understand that the NLP prompt is influenced by a lot of different factors, including the quality and quantity of the data used to train the model, the complexity of the language used in the prompt, and the specific task that the model is being asked to perform.

If you are just starting out with Generative AI, the recommendation is to start simple and get more complex as you understand how the models are reacting to your requests.

Before you start, ask yourself:

• What do you want to produce?
• What is the style or voice/tone you want to produce it in?
• What additional attributes should you consider when producing your prompt?

From there, documentation from the various applications should help guide you.

The Limits of Generative AI

While impressive in its ability to create convincing images, videos, and even text, Generative AI is riddled with limitations.

One of the main challenges is the lack of control over the output. Even with a carefully trained model, the generated content may not always align with the desired output.

In addition, generative AI may struggle with complex or abstract concepts that require nuanced understanding.

There are also huge ethical implications to the technology.

These models require a lot of expensive processing power to work, and that will result in it being inaccessible to some people and organizations.

Lastly, there is the issue of ethics.

There is legitimate concern that generative AI will be used to create fake content that could be used for disinformation or nefarious purposes, such as deepfakes.

To spot the problems, communications teams will need to be “AI-aware” because the technology is evolving at the speed of early Internet.

And, if your teams are not already talking about Generative AI, they are already behind.

-

Jennifer Johnson works in platform strategy at Horizontal Digital. She is based on Dallas. Thanks to Mike Rook for the edits!