Learning How to Speak the Same Language as Generative AI
Last week, I was working on an article about hypothetical advice a chatbot might give a person who was just appointed the 'human in the loop' at their company, and had no idea where to begin.
After 'interviewing' a number of LLMs, I wrote and revised my piece, quoting the chatbots as if they were expert spokespersons.
I then went back to Claude and ChatGPT for a critique of my draft and asked each for suggestions on what to improve.
Then, I took it a step further and got Claude and ChatGPT to incorporate their ideas directly in a rewrite. And I compared their versions to my original.
While the synthetic-infused drafts were good, they didn't really sound like me. They were missing my quirks and some of the leaps of faith I expect (hope?) my readers will take.
Sure, they flowed well but the copy was just too buttoned down and literal for my taste.
Communicating is More than Simply Connecting the Dots
It occurred to me that one of the primary reasons we have difficulty communicating with generative AI is that we're speaking two different languages. Or rather two distinct dialects of the same language: human and machine.
That makes it challenging for people and AI systems to ever be on the exact same page.
One reason is the way LLM's and people's brains are structured. While both use neurons and electricity to make connections, LLMs are designed as highly sophisticated prediction machines. They need to be trained on a lot more data (i.e. words and visuals) than us.
Also, LLMs thrive on text, not subtext. They can't read between the lines like you or me. People are constantly trying to fill in the gaps between what was said and what was implied. We turn things over in our heads, examine possibilities. We're trying to figure out nuance and hidden meanings by attempting to empathize with what the other person said.
And of course, AI systems are computer programs. They may be able to learn, but their underlying language is code. You can see how machines translate your prompts into code when you use Claude Artifacts or ChatGPT Projects to develop an app.
Build an Understanding
Learning how to have an effective two-way conversation with machines is almost like learning a new language.
Except you're starting with a strong base of knowledge (i.e. the language you speak) and you have to figure out how to modify your linguistic habits and speak 'machine'.
Here are 5 tips to get you started:
- Get to know your LLM's personality. Earlier in the month, I asked a couple of chatbots to describe their and their rivals' personalities. And the responses were pretty spot on. Before working with a model, ask yourself what the LLM's strengths are and how those match your needs. If you're looking for a pleaser, try ChatGPT or Gemini. An overthinker ... Claude. A more buttoned down corporate response - Copilot. Or, for something more off the wall, consider chatting up the cool bot, Grok.
- Be direct. Don't beat around the bush. Be as detailed and specific as you can in your instructions. Fill in the blanks. Craft your prompt so it has a more logical and well-ordered flow, with section breaks, subheads to denote the topic and bulleted or numbered lists.
- Blend structure and chat. While knowing what you want is a key element, you don't have to make the interaction all business, all the time. Start with a chat, ask the AI questions, then add in some more structured elements to your request and encourage the AI to ask you questions and provide it with more detail in your responses.
- Use delimiters, symbols machines understand. Delimiters are like punctuation marks to chatbots. They help bots make sense of your input and know when to treat a piece of text differently or to switch to another topic. Delimiters include: quotation or triple quotations, square, curly or angled brackets, hash marks to denote sections, and so on. They provide concrete steps LLsM use to process your request. Remember, you're coding with words.
- Get a gut check from another AI. Once you have your result, feed it into a second chatbot, provided you know your data is safe. Ask it to evaluate the output, say what's missing and give you another perspective. Then, use your writing and editing skills to take all the feedback, discard what you don't believe is helpful, and polish your copy.
Of course, all this takes time and patience. Quality isn't measured in productivity and speed.
A Roundup of Tools
One more thing. Learn how to get the most from tools, but always start with your ideas or first draft. In other words, keep the real creativity for you.
In this week's AI and Digital Marketing Trends video, I look at a few of the latest creative tools including Adobe Firefly's AI design app and how smoothly it integrates with your desktop for a seamless visual workflow.
I also talk about campaign reporting transparency in Google's AI-powered Performance Max campaigns. Plus a simpler way to use a prompt to edit photos in Gemini and image to video tools with TikTok's AI Alive.
Check out the video and let me know what you think.
Coming in August: New Exec Ed Course on Applied Gen AI
Before I go, I want to remind you I'll be teaching a new two-day exec education course in August: Applied Generative AI for Creativity: Writing, Research, and Multimedia Storytelling.
The two-day virtual workshop is being offered by my alma mater, the McMaster University Master of Communications Management, and it's open for registration now.
In Day 1, we'll discuss how to prompt and research more effectively by combining patience, subject matter expertise and structure.
In Day 2, we'll demo and test out various multimodal tools and think about the positive and negative implications of AI and marketing and comms.
Interested? Here's a link to the Applied AI sign up page to register or find out more. And if you use the code MARTIN50, you'll get a $50 discount!
Look forward to seeing you there!
Follow Me on LinkedIn
Well, it looks like it's the end of the period for issue #127.
Thank you to all of you who follow me and subscribe, read, comment and share this newsletter!
This newsletter comes out twice a month. But between issues, I share shorter daily posts with my take on digital marketing and the latest on generative AI. It's another way to stay on top of the trends.
Let me know if you have questions about any of the videos in Digital Marketing Trends or any of my other LinkedIn Learning courses.
You can also visit my website and send a message or a question.
How do you know you're speaking the same language as AI? Please share your thoughts in the comments below.
I hope I've made a mark 😀. See you in two weeks!
Note: All the content in this post was written by a human—me and not Martin-bot.
Great post, Martin Waxman, MCM, APR! 👏 Your point about needing to learn the language of generative AI really resonates. As you rightly highlight, AI tools are becoming more like co-pilots but with their own modes of expression, strengths, and limitations. Speaking their language means crafting precise prompts, understanding how they interpret context, and guiding them with both clarity and creativity.
Martin, the personality mapping approach particularly stands out. Each AI model has distinct communication patterns that mirror how we adapt our style for different colleagues.
Martin, does that mean AI is going to put us in a box? 😁 Metaphorically speaking, maybe we already are!
Love this take: "LLMs thrive on text, not subtexts." Important to keep in mind.
This article perfectly explains why talking to AI feels different than talking to a person. It's not just about getting the right facts, but about the unspoken feelings and hints that humans naturally pick up on. The advice about being direct and using special symbols in our prompts makes so much sense for getting AI to understand us better, while still letting our own unique writing style shine through.