I did three AI workshops last week. One was focused on AI in a world building class, one for productivity and one was focused on creativity. You can tell I’m doing too much with AI based on my ads.
All our AI presentations will start with the same pattern. This is a condensed version because it’s boring for me to type. That doesn’t mean it’s not important, I’m just easily bored.
- Fact check: LLMs are made around likely language not factual language. I use my labrador puppy metaphor.
- Privacy: Don’t put private information in LLMs. Think hard about where you put any info. Take note of the statements our two SSO-enabled LLMs provide.
- Person in the loop: Don’t send any LLM content to others without taking a look.
- What do we have? A chart provides the top five LLMs and breaks down which ones Middlebury provides access to (Gemini, Copilot). There’s a bit about their associated companies and paid/free options.
Worldbuilding
A couple hours before the class, the professor sent an email with more details and referenced creating a world for plutocratic bioluminescent snails. I found it impossible not to rewrite all my examples with this as the topic. This kind of compulsion is also why I will never be rich.
If you go through the embedded slideshow, you’ll see the progression of specificness in prompts. It was meant to lay a foundation for basic prompting while keeping things specific to the course.
You are an expert xenologist who has been living among the bioluminescent snails of Kepler-186f for 20 years. The snails have a 200-year-old plutocratic government. Using the Japanese Zaibatsu as your guide, provide a hierarchical list of the top five government positions and their responsibilities. Focus on umwelt and how communicating with light would influence their titles and society.
For me, that kind of prompt was getting close the associative trails of the days of #thoughtvectors. The difference between using Google to wander and using the LLM to wander matters but maybe less in a fictional process and maybe not at all in the near future as the AI fuses more and more with whatever searching the Internet looks likes.
The idea that I can mess around with a persona with a prompt like this is fun and interesting to me. I even made one example where I made the LLM build a blink pattern to translate back and forth since the snails couldn’t read English or hear sounds.
You are a xenologist. I am an alien bioluminescent snail. Ask me questions one at a time, to help understand my society. Focus on the government structure and understanding how we came to have a plutocracy.
Did I burn down a forest to do this? Maybe. Is this BP creating the term carbon footprint term? Maybe. I’d ask an LLM but that’s not how these things work and it’s not terribly clever. It is easy so it’ll keep on rolling despite anything.
I even wander in to some image prompts and find value even when the prompt fails utterly like this one did.
Create what a bioluminescent snail would see through 4 eyes that have evolved to perceive the finest gradients of light. It’s looking at a human spaceship landing. Consider umwelt and color spectrum perception differences.
Productivity
Far less exciting, but perhaps of interest is the productivity presentation.
Same intro, like I said. I then break up using LLMs for productivity into 5 categories.
- answer machine: gets you really specific types of answers pretty fast and very specifically (see caveats from intro)
- transforming content: OCR and table reorganizing
- content summarizing: prefect? nope but one way to condense stuff that you probably wouldn’t read anyway
- brainstorming: probably the easiest win, just a way to get rolling, the bad first draft, something to react to
- extra pair of eyes: when your significant other has been forced to read too many emails for tone . . .
Creativity
I wanted people to be able to copy the prompt and run it on their own stuff so I made this in Codepen. A chunk of the copy button code is AI-generated. The default worked well and I tweaked it a bit. Again, like the recent Flickr example just a way to get to the end faster. I didn’t come up with the idea that I needed this to work like this until an hour or so before the presentation.
These are all examples you will have seen if you’ve seen the other stuff I’ve put out in various channels. If not, they’re kind of fun.
See the Pen
non-standard AI by Tom (@twwoodward)
on CodePen.