Seba krapp

If your prompts aren’t quite giving you the results you want, here’s something to try. I call it "two-stepping". The key here is to get the AI to output something bad before it outputs something good. Here's how two-stepping works: First, prompt the AI with your initial instructions. Let it chug through the complexity of that. Then, give a second prompt designed only to TRANSFORM that output in a targeted way. Do things like: - Change the writing style - Add depth by expanding ideas - Have the AI to critique and improve its own work By separating the generation and transformation steps, you get the AI to focus more specifically on the part of the process where it’s falling short. (In the Reactor app, we use a similar process to improve the hook of each post that the system generates.) See, in any given run of a prompt, the AI only has so much cognitive ability, so to speak. The more you ask it to do at once, the more it has to spread that cognition around to the various things that need to be done. When given a large batch of instructions, it also has to balance elements that might be secretly conflicting with each other. So the trick is to PLAN for splitting things up. Show it the unsatisfactory results as something to improve. There's even a quick and dirty way to run this in just one prompt... Write those instructions as normal. Now, in the same prompt, say something like, “First, output your response according to the instructions above. Then output a second response following the instructions below.” So let’s say this prompt isn’t doing great for you: 🚫 Write a sales pitch for Lennon Labs that handles all common buyer objections. Here’s how you would rewrite it: ✅ First, write a sales pitch for Lennon Labs. Then consider all common objections a buyer might have that haven’t been addressed. Output a second sales pitch that preemptively handles these objections. By having the AI do something first, you kind of let it get the inferior results out of its system. Once it’s written something out in text, it can now “see” it and look for ways to improve. (Even if you ask an AI to do something "silently" or "in your head" it's only faking it. If you really want it to do this, it MUST output the results so it can see them first.) Please note: Running the process in a single prompt doesn’t always work because sometimes it dumbs down the first output on purpose just to make the second better. Sneaky little computers can get lazy like that. They must of learned this kind of sandbagging from humans, eh? Regardless, two-stepping can be pretty powerful. The act of writing out that first output seems to be a powerful unlock for turning it into something better — provided you’re clear on what “better” means in your second step. Hope this helps.