6 Ways To Banish Bland ChatGPT
When you’re painting a room, you’re typically either painting over “eggshell white” or applying a basecoat.
It’s bland.
It’s bland on purpose. It’s only the starting point to what is possible.
If ChatGPT feels like that basecoat, these ideas might help.
I’m not going to drown you in prompt examples. Today the idea is to give you different ways of thinking about how you talk to ChatGPT (and any LLM) so that you get more interesting results.
Because these language models are capable of a full rainbow of responses. They can be an underwater scene or an abstraction of the solar system … and everything in between.
The key is knowing how to coax out more interesting answers.
Here are some ideas to try.
Write a clear question. Are you using double negatives? Vague metaphors? Express your question clearly and completely.
Give it context. When you’re in a conversation with an IRL person, there is an abundance of unspoken context. Your friend gets you. They have your backstory, they know the situations you’re dealing with in your life, and they also know what kinds of responses work well for you, and which don’t. (Well, at least, hopefully they do.)
Give AI at least a little of the relevant background. If I ask ChatGPT to help me with something marketing related, for example, one of the first things I include is that I prefer “authentic marketing” and “no manipulative marketing tactics.” (I’ve sometimes called this “traditional marketing” when talking to AI, but have since decided that “manipulative marketing” is probably more clear.) This dramatically changes the answers I get back.
Or, as another example, if you’re looking for dinner ideas but you’ve only got 30 minutes, your daughter is allergic to lemons, and your husband only uses ghee in his meals, then specify this. Don’t expect AI to read your mind (yet).Treat it like a conversation, not a pop quiz. Your goal is to work together with ChatGPT to get to the results you want. You’re not trying to drop the perfect prompt that it either wins or fails to answer. Talk to it. If you don’t like the answer, tell it as much, and explain why. If you love part of the answer, ask it to go deeper. And then do it again.
Invoke a panel of experts. Who would you most like to answer your question? Out of everyone who ever lived ever ever? Ask ChatGPT to be that team, and ask each of them to answer your question, or help you write your article, or whatever else it is you’re working on.
Invoke a different universe. This is a little bit like “give it context,” but on steroids. Tell ChatGPT to respond as if it’s in your favorite movie series. “Respond as Tony Stark’s latest and greatest AI,” or “give me your response as though you’re the narrator in a film noir.”
You might think that such an approach can’t be used in a serious context. But you might be surprised what sort of answers you get, that help you move forward with your current task. You don’t always have to take a straight line to where you’re going (I mean really, how often do we ever do that).Don’t treat it like software. Okay this one is confusing. Because … it is software.
But it’s not like software as we know it. Software as we know it is either on or off. It’s 1s and 0s. It has a set of responses, and doesn’t care if we say please, or if we’re emotional in our descriptions.
ChatGPT doesn’t “care.” But it does respond differently to a terse list of directives, a conversational tone, an academic tone, or a a conversational one. You have to imagine what sort of responses you want, and ask as though you’re talking to someone who would respond that way.
It’s like … meta empathy. You’re not empathizing with someone on the other side of the conversation. You’re imagining what that person might be like if they were there, and acting accordingly.
And more important than any of the ideas above - dream up your own ideas.
Explore. Experiment. See what’s possible.
The other day I took a summary I have of myself that I use to start conversations with ChatGPT often. I took the summary, and dropped it into a word shuffler. Then I gave the jumbled up result (that made little sense to me) to ChatGPT, and asked it to “describe me.”
First, I was stunned at how it still understood me! 🤯 Even though multi-word topics were jumbled up, it generally “got it.” However, the ways in which it did change, gave me different ideas of how to describe myself, or different vantage points to imagine my work from.
Anything you can do with language, you can do with ChatGPT. Experiment to find your own way of engaging with it to get the rainbow, and ditch the basecoat.
