For most of my (and just about all of AI’s) life, I’ve avoided AI like the plague. I was afraid that if I ended up using it, I would be reprimanded by friends, family, and my writing peers. I was scared it would hinder my creativity, steal my dream job, and force me into something I would never be happy doing. Artificial Intelligence, for the longest time, was not smart enough to do something as complicated as take my (hopeful) career as a writer.
Recently, I’ve been using AI on a close-to-daily basis, experimenting with questions, prompts, and photos, mostly engaging my interests of sports, writing, and philosophy. I’ve been using a variety of different artificial intelligence models, including OpenAI’s ChatGPT, Google’s Gemini, and Meta’s Llama. To give an insight about some of the things I ask, here’s a small list of things I’ve said to test the creativity, direction-following, and overall intelligence of the supposed artificial intelligence.
Name me fourteen ingredients found in banana bread. Write me a limerick only using the names of MLB players without repeating names. What are genres of music similar to yacht rock. Write me a story about a rogue mercenary in a fictional cyberpunk setting.
Through some of these prompts, I’m able to get a better grasp on each unique coding behind each chat model. Some, as you’d have it, are better with creativity over problem-solving, and it was essential to use the proper one for each task.
A Dying Breed
My first concerns about using AI is that they were out to steal artist’s jobs. In fact, if you look around, you’ll most definitely see some form of AI-generated art in your daily life. As a writer (which can be associated as an artist), I was worried about it stealing my passion. I was worried about it stealing my creativity. As someone born in my day and age, there’s already so few topics one can write about that haven’t been written about in some form or another in the past.
My fears have been resolved especially after using the AI Playground through Outlier AI. I enjoy testing the creativity capacities of chatbots, especially GPT, which is known for creative writing. On writing a prompt about cyberpunk themes, I noticed that it took from other fictional cyberpunk media. It set the story in Neo-Tokyo, which I may give it a pass for, but it was mostly popularized as a cyberpunk city through the anime movie “Akira”. The “antagonist” company of the written prompt was that of “GenTech”, which loosely connects to the Cyberpunk 2077 company of “Gene-Tek”. I wanted to see if it will loosely copy other themes as well, so I had it write about a traditional western plot line.
The Wild West Conundrum
The first thing I noticed was the difference in names, but the title of the fictional town the story took place in was “Cactus Junction”. To further test the limits AI is willing to go, I asked it to create ten Wild West town names, nontraditional. It gave me names like “Gearspring Gulch”, “Copper Canyon”, and “Silver Spur”. It seems Artificial Intelligence loves alliterations.
If ChatGPT, known for its creative writing, can’t produce a Wild West town name that isn’t two words and/or alliteration, I had reservations that any other chatbot could. However, it would be no fun if I didn’t try. I do have unlimited access to the models for personal fun. Let’s choose Meta’s Llama, as it says it can do creative writing. I gave the exact same prompt: “Can you name me 10 untraditional wild west town names”. While it did give me a few two-worders and one alliteration, I received “Luminaria”, “Ravenhurst”, and “Heliosville”. It took my prompt one step above and beyond, including names tied to science-fiction, steampunk, and fantasy western settings.
I could say Meta’s Llama wins in pure creativity, but I’m still not worried about my position anymore. I decided to give one more model a chance, which is Anthropic’s Claude 3.5 Haiku. While it never stated it was great at creative writing like the other two, it said it can do “short-form creative content”. I could name some more names, but to be honest, they were eerily similar to that of Llama’s. I feel we shouldn’t have to worry about Artificial Intelligence taking any creative jobs anytime soon.
What’s Your Favorite?
I wanted to get to know some of the AI models a little bit more, so I decided to play some traditional icebreakers. Such icebreakers include, what’s your favorite (blank), two truths and a lie, and a fun fact about yourself. It’s like we’re back in college. I began with “what’s your favorite animal?”, giving the prompt to Llama, ChatGPT, and Gemini. After much back and forth (GPT refused to answer the question), I got the answer of a chameleon. Gemini, also very hard stuck on giving an answer, settled on the capybara. It was a respectable choice. Llama had no reservations about my question and AI not having personal preference, choosing the quokka. Just “Quokka”.
Now that I know a little bit about their favorite animals, let’s do “two truths and a lie”. I started with Llama, after resounding results on the first icebreaker. I needed to refine my question after getting two random facts and a lie. The question turned into “Give me two truths and a lie about you”. It was the easiest two truths and a lie I’ve ever played. GPT’s was a bit more difficult, and the lie was “I have the ability to browse the internet in real-time to find the latest information”. Still, I succeeded. Gemini tried telling me it had a personal email address to communicate with other AIs. So far, I might confidently say that AI doesn’t have the capability to lie about itself.
Lastly, I want to know a bit more about the models I’m having conversations with. Gemini claimed they can generate text in the style of different historical figures. I’ll be testing this in a future section. ChatGPT’s fun fact is that they can create creative content with unusual prompts and concepts. Their example was, a poem about a dancing toaster. Boring fun fact, may I say. Llama’s fun fact is they can understand homophonic inputs, like understanding their, they’re, and there even if it’s misspelt. I’m glad to have learned a bit more about these three bots, but now it’s time to up the ante.
An Analysis of AI Models
I’ve spent a lot of time getting to know each individual AI model listed (Llama, ChatGPT, and Gemini), and now it’s time to test some features unique to them. Going into this activity, I thought I would despise AI. It’s what I’ve been brought up to understand; AI is here to take jobs, including creatives’ jobs. Now, I fear it a lot less. Look at the Wild West town names different models created. If that’s the best competition against my future job, I’m safe for now. I’m saying all of this as someone who once avoided AI like the plague. If I contribute to asking questions, I’m contributing to making it better. I’m making it understand English in a different way and give it prompts it may have never heard of before.
Even I, as someone who swore to never use AI on creative projects, has used it lightly. Sometimes it helps to bounce ideas off of someone else, even if that person isn’t real. While I would never directly copy something from AI to the paper, I enjoyed using it to discuss ideas my friends or peers would have no desire to hear about.
Before I end this, I promised I would revisit Gemini’s promise of creating text in historical figures “voices”, so let’s see that real quickly! I’ll leave the readers of my blog post on this beautiful sonnet, written by Napoleon Bonaparte (Google’s Gemini AI).
