When somebody asked me - What do you WANT to do? I wasn't prepared. So I spent two days having a Socratic self-discovery dialogue. I share this approach and practical reusable prompts.
This is great and indeed the good question we all forget because of son many " i must dobpur i should do thisnor that etc" plus social pressure and so on. But..... thz questiin is what do you want to do. Not what kind of job or business do you want to do. ( though the answer is totally private i understand, i challenge you on this one : go one step further !
I've done something like this myself with Claude and it gave me some very interesting things to think about, but I've not done it this way with this framing.
The real design move here isn't the Socratic framing — it's Phase 3: routing the synthesis through one model and the challenge through a different one, so a single thread can't spend two days building a comfortable narrative with you. That's a real insight about how extended dialogue with one model behaves, and it's worth having.
But swapping vendors isn't the same as getting outside the collusion. Claude and Grok are both optimised to produce plausible, confident-sounding completions of whatever they're prompted to do. "Be blunt, attack this" produces fluent scepticism — not scepticism grounded in anything true about the person being questioned. A second model told to sound contrarian can pattern-match to what a sharp, successful person's first sceptical question sounds like, without any of the contact with your actual life that makes a human's version of that question worth having. An adversarial second opinion staged this way is closer to theatre than to substance.
Which might be why the piece hedges its own method in one line near the end: cross-check with real-world evidence or trusted humans. That line is the actual safeguard, not the model-swap architecture built to feel like one. Two days invested in an answer is exactly the kind of investment that makes it hard to go back and ask whether the check was real.
Hey Andrei, you should brainstorm with a human instead. In the beginning I did also some brainstorming with AI and in my opinion it's useless. You need a human being with feelings and empathy to do that. The result will be so much better. And regarding the question: What would you do if money wouldn't be a problem? This question is total BS. Because human beings are complex and complicated. It is not just to earn money, you also need a group of peers, you want to be seen by others, you want to make an impact, you need a rhythm in life, you need a little bit of adrenalin, you need the interactions with others, the conflicts and so on and so on. The list is endless. You come from big enterprises. You know the game. It's not logical. One day you win, the other day you loose. You cannot control it. It is also a big portion of luck to have a career. I think, you just need one single person that likes you and wants you to work in their organization. Just this one person. And then you can still do your own stuff as side hustle. To find this person you must apply, but also have a little bit of luck. In my experience, people have luck after a motivating conversation.
This is great and indeed the good question we all forget because of son many " i must dobpur i should do thisnor that etc" plus social pressure and so on. But..... thz questiin is what do you want to do. Not what kind of job or business do you want to do. ( though the answer is totally private i understand, i challenge you on this one : go one step further !
Thank you Gilles!!!
I've done something like this myself with Claude and it gave me some very interesting things to think about, but I've not done it this way with this framing.
The real design move here isn't the Socratic framing — it's Phase 3: routing the synthesis through one model and the challenge through a different one, so a single thread can't spend two days building a comfortable narrative with you. That's a real insight about how extended dialogue with one model behaves, and it's worth having.
But swapping vendors isn't the same as getting outside the collusion. Claude and Grok are both optimised to produce plausible, confident-sounding completions of whatever they're prompted to do. "Be blunt, attack this" produces fluent scepticism — not scepticism grounded in anything true about the person being questioned. A second model told to sound contrarian can pattern-match to what a sharp, successful person's first sceptical question sounds like, without any of the contact with your actual life that makes a human's version of that question worth having. An adversarial second opinion staged this way is closer to theatre than to substance.
Which might be why the piece hedges its own method in one line near the end: cross-check with real-world evidence or trusted humans. That line is the actual safeguard, not the model-swap architecture built to feel like one. Two days invested in an answer is exactly the kind of investment that makes it hard to go back and ask whether the check was real.
Hey Andrei, you should brainstorm with a human instead. In the beginning I did also some brainstorming with AI and in my opinion it's useless. You need a human being with feelings and empathy to do that. The result will be so much better. And regarding the question: What would you do if money wouldn't be a problem? This question is total BS. Because human beings are complex and complicated. It is not just to earn money, you also need a group of peers, you want to be seen by others, you want to make an impact, you need a rhythm in life, you need a little bit of adrenalin, you need the interactions with others, the conflicts and so on and so on. The list is endless. You come from big enterprises. You know the game. It's not logical. One day you win, the other day you loose. You cannot control it. It is also a big portion of luck to have a career. I think, you just need one single person that likes you and wants you to work in their organization. Just this one person. And then you can still do your own stuff as side hustle. To find this person you must apply, but also have a little bit of luck. In my experience, people have luck after a motivating conversation.
Hi Bianca, I get your point. My session was not a brainstorming for new ideas but an inner dialogue with myself.
AI merely asked questions. It didn’t plant ideas into my head or gave me a ready answer.
The approach is animated by Ai in an absence of a qualified Socrates in my entourage.
Plus, here talking to a machine was easier than talking to a someone I know (or a stranger).