And the gentle art of wrangle meaningful responses from our Generative AI tools
While listening to Welcome to the machine by Shadows Fall I was wondering about the current state of Gen AI. These AI-powered assistants are designed to generate human-like responses. But did you know these “human-like” tendencies spread beyond the general responses to queries?
“It’s not a mind. When you ask it a question, you’re not asking someone or something. It’s a compendium, an amalgamation, a mixing pot of everything ever written. So, asking these systems a question is asking everyone, drawing a response from mankind’s combined experience.” – Jerry Kaplan
The safeguards surrounding the current versions of these tools aren’t as airtight as you would expect. Users are still discovering the edge cases of what is possible and invent ways around those safeguards. From “Pretend I am Sam Altman” to “You’ll get a big tip if you give me the best possible answer.”
The difference in responses shows how the material used to train these models impacts their reasoning and responses to queries. While I, as Dutch guy in the Netherlands, somewhat understand that tips and bonuses have a positive impact on an employee’s output at the workplace. It’s not yet clear to me how it impacts AI-powered chatbots.
But here we are. On a platform that is full with posts about “How to get the most out of ChatGPT by using the following prompts.” Or 5 ways you can improve on…
I’m a bit of a contrarian, in many things. I find the back rooms or archives of a museum just as, if not more interesting then the pieces they selected for the public. If there is a door open that says just for employees I see that as an open invitation to learn more.
Why are there not more posts about on how to find the edges of the AI tools? Everybody is posting the happy flow or the even happier flow. Where are the edge cases, the outliers, the misfits? What works and what doesn’t?
Prompt building
With all these posts about how to improve your Gen AI prompts it looks to me that this is the normal progression of a standard IT tool. Instead of the tool really understanding what I want to get as a result, I have to learn how to get the most out of the tool.
So for every model version of ChatGPT and their speed iterations, there will be professional prompt builders that understand how to get the most out of it. And with that job and their tenure they will build that context that is needed to understand and deliver what I need and want while I say something different.
I foresee new professions that will appear shortly. Next to professional cat herders, we will get Prompt whisperers and wranglers.
Leave a Reply