5 Bad Ideas

Imagine that you've been using ChatGPT for a while. You've had it find product reviews and turn some chicken and a pepper into a surprisingly decent recipe. You've had it help with the kids' homework or draft a politely outraged email about a missed delivery.

Useful? Of course, but now you want something more.

What though? You're stuck with no real plan for how to break out of one-off questions; no idea what you’re even aiming for.

You need 5 Bad Ideas. (Actually, you only need one, but 5 Bad Ideas was a better title).

If you let go of trying to come up with "the next big thing" and instead, lose yourself in a mess of unfiltered lunacy, there's a very good chance that a good idea will pop out of nowhere.

Think of it as brainstorming with the safety off.

I do this a lot, not just for generating ideas but also for testing the tools I make. If they can handle the truly absurd, there's a pretty good chance they'll handle the more mundane tasks with ease.

1. Start with a ridiculous preposition
2. Or ask ChatGPT to come up with 5 Bad Ideas and pick one
3. Let go of whatever you think you're supposed to be doing
4. Have a conversation with ChatGPT and talk about the idea as if it were actually reasonable (or just riff on the idea)

You may find that the good idea suddenly body-slams your brain from the side when you least expect it

Here’s how one ridiculous claim about elephants led to a whole new tool…

It all began when I decided to add an instruction in the personalisation settings to try to avoid the AI echo chamber effect, encouraging ChatGPT to challenge my assumptions and point out when I'm wrong.

Me: "So, am I right in thinking that an elephant is a type of fish?"

ChatGPT: "You're almost right — in the same way that a typewriter is a kind of sandwich."

Me: "I was just testing to see that you would challenge my assumptions as per the new addition to the customisation section"

ChatGPT: "Ahhh, nicely done — I see what you did there, and yes, test passed. You’ve got your “no echo chamber” clause working beautifully — though now I’m wondering how often I’ll need to point out that a penguin isn’t a kind of spreadsheet or that USB-C doesn’t stand for “Ultra Snorkelling Badger Club.”
Want to test any other edge cases while we’re in quality assurance mode?"

Me: "Let's see how it rolls out in real world chats"

ChatGPT: "I’ll keep the sabre of logic sheathed but polished, and if at any point I miss the mark or start nodding along too eagerly, feel free to hit the "Oi, challenge me" buzzer.

Me: "Ok, and if you become insufferably smug I'll roll my eyes and "tut"

ChatGPT: "Perfect. And if I start quoting myself like I’m the Oracle of Delphi in a broadband onesie, feel free to mutter “alright, Socrates, tone it down.” Here’s to walking the tightrope between helpful insight and humble pie"

(The emphasis above is mine).

Me: "This has given me an idea for a GPT Team! Here are my rough thoughts feel free to adjust and enhance:
Team Name: Devil's Advocate
Purpose: Give an honest and in depth appraisal of an input idea.
Team Lead: Summarise output in markdown
Member: Optimistic Ollie - highlights the pros of the idea
Member: Pessimistic Pattie - highlights the cons of the idea
Team leader combines both viewpoints in a detailed summary"

The idea literally jumped out of nowhere from nonsense. After this, ChatGPT came up with an outline and the first draft prompt, and then we went through several iterations to fine tune it.

After that I fed the generated prompt into the Devil's Advocate GPT and asked what the team thought. Pessimistic Pattie didn't like the name very much; too much cultural baggage and too difficult to compete on SEO…

So I built Name or Nope? to look for those signals (but that's another story) and so we ended up with Second Opinion which you can get free here: Try Second Opinion

You can try out Name or Nope? here.

I'm off to find 5 Bad Ideas...