How Google AI course helped us with designing wedding invite cards

Though I use AI nearly every day, I decided to pass a prompting course. The logic is simple - corporations who developed AI know how to use it at maximum, so it makes sense to learn from them. I took the Google authored prompting course on Coursera, and here is what I think about it... But hey, Sport, is that a Far Cry Blood Dragon style wedding in the title?!

New language

Fair indeed - AI+LLM is a new way to communicate with a computer. Previously there were only programmers to tame the machine. But today mortals too can bend it to their will by talking to it in human language and get results which were not exactly programmed. I never thought about it, but the course makes a strong point here.

Racism, ageism and biases in AI, really?

The most unexpected item in the course for me was a warning that I have to write my prompts in a tolerant and inclusive way, as AI can pick up the tone and response with stereotypes.

Yes, AI can make mistakes, but Google does a serious accent on it. It's a surprising honesty in the course that's supposed to promote AI into the masses, and I guess it deserves respect.

Framework and practice

The course gives good examples of how to formulate prompts to make AI understand you correctly. While for me there was not much new (I studied as a robotics engineer, so I had AI course in university), it definitely worth learning, as prompting has more structure that most people think. Adding constraints, describing context, setting references - these all things float down there.

A word of warning though: some AIs may not respond well to the approach described by Google. I'd say, by my experience, Copilot, ChatGPT and Gemini are flexible and understanding to the user's request, adjusting preferences to their way of thinking easily. But others (I can think of Meta AI here) are way more rigid.

Even though I use AI every day, there were a couple of examples with new ideas. Let's say asking AI to impersonate someone and talk to them as a rehearsal - that's an interesting way of use! Never thought about it before, but now I do.

Wedding cards

As you might suggest, making wedding invitation cards is a thing - you don't have space for mistakes. I'm, however, not a visual designer, and we came to this item too late. We could run around with experts, but I am a certified prompting specialist, so...

First, I described what I want and in which visual style (context). We had a Greek style wedding, so the prompt was:

generate a wedding invitation card, in greek style (blue, white and golden colors), keep some white space in the middle of the card for the text of the invitation

The first result was so-so:

What is this?! Yes, I could add constraints from the very beginning, but who knew that the AI would add a helmet and columns on an invitation card?! I could iterate, but weaponized by knowledge I didn't. Instead, I asked the AI to remove the elements I didn't want:

That looks neat! 10 mins of AI chatting and evaluation, and here we are - only had to write the text and print them. And... yeah, not just us, but the guests appreciated the cards highly.

Verdict

So, guys, even if you're an experienced AI chatter, I still recommend passing a prompting course, especially, considering that it takes just several hours. The structure and examples that it gives help to bend AI to your will wa-a-ay faster and easier.

What the course missed

To my surprise, the course doesn't talk about request formulation styles (which I guess any engineer learned on university AI lectures), so I'll fill the gap:

• Imperative (how to achieve the result): you step by step describe to AI what to do. For example, "I'll ask you several questions about life situations. Explain me how a true samurai would solve them. Then turn the result into antonym - instead of courageous actions propose cowardly ones, turn loyalty into betrayal and so on." Imperative approach is good when you need precise results or a complex chain of actions.
• Declarative (what result to return): you describe only the form of the result, and AI decides how to achieve it by itself. For example, "I'll ask you several questions about life situations. Tell me how a samurai would solve them. Give me the result in the form of a haiku of 5-7-5 size." Declarative style fits better when you're preparing formal things: meeting notes, follow-ups, agenda, etc. And as you're describing the form of the results, then if you have an example of it, it's a good idea to give it to AI. E.g. give it 3 existing product description examples before asking to generate a description for a new product.