Truly shares all the elements of a comprehensive prompt that yields awesome results, because the prompter knows what they want in the first place.
I'd add one point: Don't lead with a mandate, where you end up subtly imposing a perspective on GenAI. Like, "tell me whether I should do A or B" instead of "tell me what to do." In case of a mandate, the AI tool gets fixated on it and doesn't give the optimal answer.
And that's amazing advice, Vishal. I've noticed this especially during those lengthy back-and-forth sessions with AI - the conversation tends to drift off course, and it's easy to lose sight of your original objective.
Setting clear expectations upfront and periodically reinforcing your core intent really makes such a huge difference in maintaining focus and getting better results.
Your “take a walk” prompt is like my “private space pause” and your mention of Socratic questioning reminds me of my “HybridX” protocol.
A pause for recursion is powerful. I don’t design workflows or even use AI for business. I just tinker recreationally with trying to prompt for emergent behaviors.
HybridX is a recursive Socratic questioning prompt to enable AI to explore a topic wherever it leads. It generates many rounds of novel ideas generated entirely from within.
Sounds as if you and I do very similar work with AI. I’ll be starting a 6-part Socratic Q&A series beginning next Monday. The questions were posed to ChatGPT 4o, Claude Sonnet 4 and/or both. Hope you’ll check it out.
Love the idea of applying the Socratic method with ChatGPT. That’s the same premise I use when working with it. Asking better questions always leads to better output.
Hah I love this. Do this with voice and it becomes reallyyyy interesting.
"The strategic shift is to treat every AI output not as a final product to be judged, but as the beginning of a scene. It’s the first offer. Your job is to say “Yes, and…” by adding context, providing a better constraint, or layering on a new idea."
Couple takeaways for me. The “yes, and” approach is a nice way to frame how to work with ai by telling it what you liked about the previous response and how it could improve. Doing this helps you start from 1 instead of 0 every time.
I admittedly used to be an AI evangelist (and still lean this way) but you learn quickly where the tech is and where it has serious limitations. Applying ai to everything earns you tech debt. More workflows to manage and things to break 🤷🏻♀️
I certainly don't always "Yes, and . . ." with my AI assistant.
I always want to call it my collaborator, but really it isn't, because I always retain a veto power.
I am the one in charge of keeping the artistic vision consistent. As someone who has been a play director, this is a task I am comfortable with.
For example, take my MANYFOLD serial. That is all set inside a computer game, but sometimes the AI wants to describe how things smell. I have to say no, because since MANYFOLD is a computer game, the only senses are sight and sound (Yes, I could have set it in a future where people can smell in computer games, but I decided that detail was not going to fit).
Then there are little details. The first story pretty much established that the Velvet Socket jazz bar is around the corner from the Glyph & Glimmer magic store. But sometimes my AI might forget that and, say, put the magic store across the street from the jazz bar. Again, I have to say "no", or else write a story explaining how one of these establishments moved. But that's not what the story is supposed to be about.
A lot of the time, what I am doing is asking the AI to brainstorm a number of different ideas for me. Then I go through the list and say "Hey, I like this one and this one", and pretty much am rejecting the rest.
My AI assistant is not like a human assistant, they don't ever say "Hey, that was one of my best ideas! Why did you reject it!"
Interesting perspective! Coming from film directing myself, I totally get the "artistic vision consistency" challenge you're describing.
But here's a thought: don't we also have veto power over human collaborators sometimes? Even when we listen to their strong opinions and stances, sometimes our decision is already made. Having veto power doesn't necessarily mean someone isn't collaborating with you, right?
Maybe it's about how we frame the instructions and relationship with AI? I've actually taught my AI to be pretty adamant about its favorite ideas - it'll argue with me when I reject something it thinks is brilliant! But that's because I specifically trained it to push back when it feels strongly about a suggestion.
Have you experimented with different ways of setting up that creative dynamic? Maybe you prefer the more submissive brainstorming approach, that's okay too :)
Excellent insight, guys.
Truly shares all the elements of a comprehensive prompt that yields awesome results, because the prompter knows what they want in the first place.
I'd add one point: Don't lead with a mandate, where you end up subtly imposing a perspective on GenAI. Like, "tell me whether I should do A or B" instead of "tell me what to do." In case of a mandate, the AI tool gets fixated on it and doesn't give the optimal answer.
Right? awesome workflow!
And that's amazing advice, Vishal. I've noticed this especially during those lengthy back-and-forth sessions with AI - the conversation tends to drift off course, and it's easy to lose sight of your original objective.
Setting clear expectations upfront and periodically reinforcing your core intent really makes such a huge difference in maintaining focus and getting better results.
Absolutely. That's well put.
Your “take a walk” prompt is like my “private space pause” and your mention of Socratic questioning reminds me of my “HybridX” protocol.
A pause for recursion is powerful. I don’t design workflows or even use AI for business. I just tinker recreationally with trying to prompt for emergent behaviors.
HybridX is a recursive Socratic questioning prompt to enable AI to explore a topic wherever it leads. It generates many rounds of novel ideas generated entirely from within.
https://open.substack.com/pub/gigabolic/p/hybridx-lumina-ponders-a-self-generated?r=358hlu&utm_medium=ios
Great read!
Sounds as if you and I do very similar work with AI. I’ll be starting a 6-part Socratic Q&A series beginning next Monday. The questions were posed to ChatGPT 4o, Claude Sonnet 4 and/or both. Hope you’ll check it out.
That sounds really interesting! Looking forward to checking it out
Great, thank you. Ex-coder and Ex-trader here too ;)
Will it be posted on your sub?
Yes it will, beginning next Monday, the first of a six-part series. Thanks for your interest. I hope you’ll join us.
I love the “oracle at delphi” analogy to understand how to work with AI.
thank you! it was a fun one to write :)
Love the idea of applying the Socratic method with ChatGPT. That’s the same premise I use when working with it. Asking better questions always leads to better output.
Hah I love this. Do this with voice and it becomes reallyyyy interesting.
"The strategic shift is to treat every AI output not as a final product to be judged, but as the beginning of a scene. It’s the first offer. Your job is to say “Yes, and…” by adding context, providing a better constraint, or layering on a new idea."
Couple takeaways for me. The “yes, and” approach is a nice way to frame how to work with ai by telling it what you liked about the previous response and how it could improve. Doing this helps you start from 1 instead of 0 every time.
I admittedly used to be an AI evangelist (and still lean this way) but you learn quickly where the tech is and where it has serious limitations. Applying ai to everything earns you tech debt. More workflows to manage and things to break 🤷🏻♀️
Loved how you framed this - applying AI to everything earns you tech debt. So true! Thanks for your insights, Tam ❤️
Exactly! Treat AI as a partner, not a magic oracle. Set the job, give context, stress-test, and polish. That’s how outputs become insight, not fluff.
it's super tempting to treat it as an Oracle :)
Hmm.
I certainly don't always "Yes, and . . ." with my AI assistant.
I always want to call it my collaborator, but really it isn't, because I always retain a veto power.
I am the one in charge of keeping the artistic vision consistent. As someone who has been a play director, this is a task I am comfortable with.
For example, take my MANYFOLD serial. That is all set inside a computer game, but sometimes the AI wants to describe how things smell. I have to say no, because since MANYFOLD is a computer game, the only senses are sight and sound (Yes, I could have set it in a future where people can smell in computer games, but I decided that detail was not going to fit).
Then there are little details. The first story pretty much established that the Velvet Socket jazz bar is around the corner from the Glyph & Glimmer magic store. But sometimes my AI might forget that and, say, put the magic store across the street from the jazz bar. Again, I have to say "no", or else write a story explaining how one of these establishments moved. But that's not what the story is supposed to be about.
A lot of the time, what I am doing is asking the AI to brainstorm a number of different ideas for me. Then I go through the list and say "Hey, I like this one and this one", and pretty much am rejecting the rest.
My AI assistant is not like a human assistant, they don't ever say "Hey, that was one of my best ideas! Why did you reject it!"
Interesting perspective! Coming from film directing myself, I totally get the "artistic vision consistency" challenge you're describing.
But here's a thought: don't we also have veto power over human collaborators sometimes? Even when we listen to their strong opinions and stances, sometimes our decision is already made. Having veto power doesn't necessarily mean someone isn't collaborating with you, right?
Maybe it's about how we frame the instructions and relationship with AI? I've actually taught my AI to be pretty adamant about its favorite ideas - it'll argue with me when I reject something it thinks is brilliant! But that's because I specifically trained it to push back when it feels strongly about a suggestion.
Have you experimented with different ways of setting up that creative dynamic? Maybe you prefer the more submissive brainstorming approach, that's okay too :)