Crush Your Presentation with AI-generated Slides
Giving a presentation on AI? Don’t bore them with bullet points. Instead, use AI to build your presentation and showcase your expertise.
Welcome to Episode 31 in Prolego’s Generative AI series. I recently spoke at Savannah’s Annual GRIT Conference about AI, and I used the Dall-E 3 model within ChatGPT to generate the images for my slides. Here are some examples.
These slides made my presentation more visually compelling, and I used the creation process to reinforce my points about AI. I have no design skills and little experience with image generation. ChatGPT Plus and Powerpoint were my only tools.
After writing a speech outline, I created a slide design style. Most AI imagery is boring: everyone makes a picture of a brain, a creepy robot, or a network. So I began by asking ChatGPT to generate two different design styles, and I picked one that sounded more human and unique.
I then created a new GPT and pasted the design style into the instructions. This approach allowed me to reuse the style for every image. Then I started generating images based on my speech outline. For example, I described a scene of me walking to my first AI class on Stanford’s campus in 1994.
The images looked ok, but they didn’t really excite me. Then I created this image with the prompt, “neurophysiologist Warren McCulloch and mathematician Walter Pitts writing a paper about neural networks in 1943.”
I absolutely loved this style. I don’t have the design skills to describe why, it just looked very human to me. So I took this image and asked ChatGPT to generate a new design style based on it. ChatGPT called this style Intimate and Timeless Craftsmanship, and I pasted it into the GPT instructions.
At this point it was just a matter of trying different prompts to get an image that captured the key points I wanted to convey, such as me at Stanford making my first neural network in 1994, and a high-performing startup team.
A few lessons-learned. First, like everything in AI, getting something is easy. Getting what you want is considerably harder. It took trial and error to get a set of workable slides, and I stopped when I ran out of time.
In some cases I changed my speech when I couldn’t get the image to work. For example I tried for hours to generate an image with a tennis ball stuck to a ceiling. No matter how I wrote the prompts, it always created an image with a skylight, and would even insist that no skylight existed. Eventually I just changed my speech to include a ball stuck to the wall.
Finally, it was challenging to generate images that reflect modern values of respect and inclusion. My design style caused the model to generate mostly images of white men from the late 19th century — I wanted people who looked like the cast of Hamilton. This took some extra effort.
I incorporated these AI lessons into my speech, and it helped explain concepts like prompt system design, hallucinations, and skills of the future. With a bit of effort, you can do the same and crush your next presentation.
