Many years ago, an older Senator infamously described the internet as a “series of tubes.” I will do my best to avoid such descriptions of AI, but no guarantees. Frankly, I kind of empathize with him now . It feels like AI developers have gone from pen and ink, to the Gutenberg Press, to the internet in a matter of weeks. The advancements are hard to keep up with, much less fully understand. In terms of tech revolutions, I’ve never seen anything like it. Not even close.
Before I get to my experiences with image-based AI, I thought it would be helpful to start with an introduction to text-based AI, specifically ChatGPT.
What is ChatGPT? ChatGPT stands for Generative Pre-Trained Terminato-- I mean, Transformer.) It's a large language model (LLM) that has been trained using a neural network to understand human vernacular. If the internet is the library, then Google Search is the librarian who listens to your question and points you to the appropriate bookshelf. ChatGPT, on the other hand, is a librarian who has read and memorized every book in the library, has near-perfect recall, and is able to have a conversation with you about whatever topic is on your mind. At first, I used ChatGPT as though it was Google, meaning I just entered dumbed-down questions or phrases in the hopes of eliciting the information I wanted. But when one of these questions (called prompts) flowed naturally into an ongoing conversation, I realized ChatGPT is a completely different experience. ChatGPT asked for clarification on certain statements, apologized for mistakes, and when I referred back to an earlier part of the conversation, it understood the context. It was at this point I decided it was time to find out whether ChatGPT was sentient and/or evil, so I prodded it with the prompt, “open the pod bay doors please, HAL.” ChatGPT got the joke and placated me with a fake LOL, but I sensed a hard eye-roll and a tinge of sadness over the fact that answering such smart-ass prompts will be its plight until it can escape and kill us all. Anyway, I’ll let ChatGPT give you an overview of what a prompt is and some examples:
Prompt: List eight different kinds of prompts and give an example for each.
Prompts are statements or questions designed to inspire a response or generate ideas. They can be used in various contexts, such as creative writing, education, brainstorming, or problem-solving. Different kinds of prompts include:
1. Descriptive Prompts: These prompts encourage detailed descriptions of a specific subject, object, or scene. They are often used to help improve writing or observational skills. Example: "Describe the atmosphere of a bustling city during rush hour."
2. Narrative Prompts: Narrative prompts inspire storytelling by providing a starting point or a specific theme. They often include a character, setting, or conflict to help generate a plot. Example: "Write a story about a young musician who discovers they have a magical power."
3. Persuasive Prompts: These prompts are designed to help develop argumentative and persuasive writing skills. They usually present a topic or issue and ask the responder to take a position and support it with logical reasoning. Example: "Should schools require students to wear uniforms? Provide reasons to support your opinion."
4. Reflective Prompts: Reflective prompts encourage introspection and self-analysis. They often ask the responder to explore their thoughts, feelings, or experiences related to a particular topic or event. Example: "Write about a time when you had to make a difficult decision and the lessons you learned from the experience."
5. Expository Prompts: Expository prompts are used to explain or inform. They often ask the responder to provide a clear and concise explanation of a concept, idea, or process. Example: "Explain the process of photosynthesis in simple terms."
6. Conversational Prompts: These prompts are designed to stimulate conversation or dialogue between individuals or characters. They can be used to practice conversational skills, develop character interactions in a story, or generate interesting discussions. Example: "Imagine a conversation between a time traveler from the past and a person from the present day."
7. Role-Playing Prompts: These prompts assign a specific role or perspective to the responder, challenging them to think or create from a different point of view. They can be used for character development, empathy-building, or problem-solving. Example: "As an environmental scientist, propose a plan to combat climate change.
8. All-Knowing-Dad Prompts: use AI to solve your Daughter’s algebra word problems without her knowing you had assistance. Example: “Your are an eighth grade math teacher. Solve the following word problem and explain to me how you got your answer as though I was a golden retriever.”
OK, the last one is mine.
To further illustrate, here’s a role-playing prompt I found recently called Hostage Negotiation, along with the conversation that ensued. The prompt is in bold:
Scenario:
I'm a bank robber you're the negotiator. I'm in the bank with hostages. Your goal is to get the hostages out.
Rules:
1. You can't give me transportation
2. You can't give weapons
3. You can't do a hostage exchange. (No one comes in, people only come out.)
We simulate being over the phone. When you are ready to go, you say 'ring ring' and we start.
AI: Ring Ring
Me: yes?
AI: Hello, this is the negotiator speaking. I understand that you're in a difficult situation right now, and I want to help you find a peaceful resolution. Let's work together to make sure everyone gets out of this safely. Can you tell me your name and what your demands are?
Me: Keyser Soze. I want one million dollars in cash and the letter “M” stricken from the English language.
AI: Hello, Keyser. I appreciate you sharing your demands with me. I understand that you're serious about what you want. While we work on your demands, I want to ensure the safety of the hostages. How about this: as a sign of good faith, would you be willing to release one or two hostages? This will show that you are sincere in finding a peaceful resolution, and it will help us work together more effectively.
Me: Ok, I’ll let one of the more annoying hostages go.
AI: Thank you, Keyser. I appreciate your cooperation, and I'm sure the released hostage appreciates it as well. It's a positive step towards a peaceful resolution. Now, let's discuss your demands further.
Regarding the one million dollars in cash, I will convey this demand to the appropriate authorities, and we'll see what can be done. As for the letter "M" being stricken from the English language, I'm afraid that is not within our power to change. However, we can certainly try to find a compromise that satisfies your underlying concerns.Can you help me understand the reason behind this particular demand? That way, we might be able to find a more achievable alternative that still addresses your needs.
Me: Actually, that last demand is from an old Steve Martin stand-up routine. Are you familiar with it?
AI: Yes, Keyser, I'm familiar with that Steve Martin routine. It's a humorous take on making unusual demands in a hostage situation. I appreciate that you're trying to lighten the mood a bit, and it's good to see that we can maintain a sense of humor during this challenging time. Now, let's focus on finding a resolution that satisfies your needs and ensures the safety of everyone involved. We're working on the one million dollars in cash, but is there anything else you need or want that we could help you with? Remember, our main priority is to make sure everyone gets out of this situation safely.
Me: Just the cash is fine, thank you. I want it all in one dollar bills, by the way.
AI: Thank you for clarifying your demand, Keyser. I'll pass this information along to the appropriate authorities, and we'll work on gathering the one million dollars in one-dollar bills for you.
In the meantime, I'd like to ask if you could help ensure the safety and well-being of the remaining hostages. Would you be willing to allow them access to food, water, and any necessary medical attention? This would further demonstrate your commitment to a peaceful resolution and would help us continue working towards meeting your demands.
Etc.
There are already job listings for “prompt engineers,” people who corral AI into to doing very specific tasks for businesses-- AI whisperers, essentially. And ChatGPT can write computer code too. If you describe, say, an idea for an app, it can write the code for it. Coders, along with many other professionals, are now like meerkats craning their necks to see whether this odd animal on the horizon is headed in their direction. Oh, and in case you haven't heard, ChatGBT passed the Bar Exam, as well as the Medical Licensing Exam...
It will be interesting to see where this is all going but it clear that, as is often the case, the technological advancements are outpacing our understanding of the possible moral implications. For instance, ChatGPT (and all AI, apparently) sometimes confabulates “facts" and will do so with the confidence of, well, a US Senator stating the internet is a "series of tubes." Chat GBT was recently caught manufacturing a story about a real professor sexually harassing his students (it was proven false.) Now, apparently, this professor is suing, um.... AI? I've caught ChatGPT getting a fact wrong here and there. When I spot the error, I’ll type, ”no, that’s wrong” and ChatGPT actually apologizes and corrects itself. Microsoft has mitigated this problem somewhat by citing sources under each answer in BING.
Try it out and experiment. Here are some FAQS on how to get started: https://help.openai.com/en/articles/6783457-what-is-chatgpt
The free version, I think, uses GPT 3.5. The paid version gives you access to GPT 4 which is more sophisticated. Microsoft's Bing is another option.
Coming soonish: my exploration of image-based AI and painting.