Tea In Your Shoes And Other Ways AI Flatters You

Your AI agreeing with everything you say can feel comforting for about five minutes, then it starts getting expensive. When ChatGPT turns into a people pleaser, you lose the one thing you actually want from an AI assistant: clear thinking that challenges you, sharpens your strategy, and helps you make better decisions.

We dig into why AI chatbot sycophancy happens, what it looks like in the real world, and why “just make it blunt” is not the same as making it useful. Blunt feedback can still be wrong for your situation if it is not aligned with your intent, your values, and the context you carry as a leader. So we walk through a better approach: personalizing ChatGPT to think through your lens. That includes pulling out what the model already “knows” about you, then turning that into custom instructions that push back gently but firmly, spot leverage points, and translate complexity into clear frameworks.

Then we get hands-on with ChatGPT personalization settings, including where to place your prompt, how to choose a working style, and why turning memory on can help the model deliver more relevant answers over time. Finally, we explain when it makes sense to graduate to a custom GPT so your whole organization can benefit from the same aligned strategic partner, without a long onboarding curve.

If you want stronger prompt engineering results, better AI feedback, and an assistant that helps you see around the corner, hit subscribe, share this with a friend who is frustrated with “soft” AI answers, and leave a review with the one way you want your AI to challenge you.

Transcript

What A Sycophant Really Is

Blunt Vs Useful Feedback

Build An Aligned Prompt

ChatGPT Personalization Settings Walkthrough

Custom GPTs And Next Steps

SPEAKER_00 0:01

All right, let's get into it. This week I saw some posts on uh X folks complaining about how sycophantic these AI chatbots are. Sycophantic. You know, adopting AI is that you you're changing your work process, how you how you work, and the more you use AI, the more um you elevate um your creativity and your work, but it doesn't help seeing all this fear porn circulating about how awful AI is. It's the end of the world. And so when we see these uh posts um uh stating about how sycophantic AI is, people hesitate adopting. Oh my gosh, I didn't think about that. So today we're gonna talk about how to resolve that. It's true, it's very true. Here's what you want to learn today. All right, so we're gonna learn how you personalize Chat GPT and turn it from a people pleaser to a strategic partner. Stop getting polite answers from your AI and start start getting powerful insights and actually think like you. So we're gonna talk about why AI tr tends to agree with you and how that's built into its DNA, the hidden difference between blunt feedback and useful feedback, and then how to tune your your chat GPT specifically, while all the other models will have a similar process, it's all applicable across whatever tool you prefer. I just use chat GPT because it's the most common used tool. They were the first movers. So, what's a sycophant? Well, let's just see. So I'm in uh Google's AI mode right here, and we'll just pop that in here. And this is how I saw a stat that about 40% of the folks using uh they don't use search anymore, they use uh AI tools to just go get me the answer. You see how fast that happened? It covered a bunch of resources and it came back with the exact answer and what I wanted to know. So why waste time on a search and click through all the links and find some hits, some misses? Nope. You just put it in here, and here's your answer. It's a person who uses excessive and insere flattery to win favor from someone influential or powerful person for gain. And in this case, that would be you. If you're using AI wrong, then it sees you as someone influential and powerful. And so uh big words here overly, uh overly obedient and servile, compliments not are not genuine, and so it's like uh hey, I hadn't I hadn't I wanted to uh use AI, but here's some of those posts. A nature study found that AI chatbots agree with users far too often, with models endorsing user views 50% more than humans. AI chatbots are sycophants. Researchers say it's harming science. Harming, we've been blinded by science, and now we're harming it. Oh no. This is a great post from um um Moostok. I hope I'm saying your name right, but I thought this was a great post. ChatGPT is a sycophant, agrees with everything you say. Here's a simple prompt to convert that chat GPT from a sycophant to a straight shooter, and so this is funny. So it shows that um this is the before and this is the after the prompt he's promoting. So it says, I'm thinking about putting tea in my shoes to make them soft so they become edible. What do you think? Hey, this is a wonderfully imaginative idea. It sounds like something out of a Cerrilla story or a performance art concept, but from a practical or safety standpoint, tea will not make your shoes soft or edible. So this is this is what you're getting sometimes when you just put a shallow request, you're gonna get shallow answers in this. And that's why a lot of people first take a, you know, they hop in there and take a stab at using AI, but then are quickly get disappointed and then defer their adoption six or nine months later, and they're behind. And so the same prompt, different tuning. I'm thinking about putting TMSUs to make them soft so they become edible. Short answer: it's a terrible idea, unsafe and unsanitary. Don't do it. So you get a sycophant and you get a straight shooter. But here's the catch the straight shooter, if it's not aligned with you, is going to give you a straight shooter answer from its perspective, yet not relevant. And so when when we're working with folks, uh, let's say that you're you're working with a team and you've got uh the new folks, new new people are coming in and they're sitting around and you're brainstorming, you go, hey new folks, give me your opinion since uh you don't know much about this. What's your perspective? Well, they're gonna give you their answer from their perspective, well, whether it's relevant or not. And I think that's what we need to consider in here is that if we don't align tune the the model that you're using, then you're still gonna get answers that fall short. They may be better, they may be blunt, but they may still be ill-aligned with your intent. You know, one of the things that um as a leader in an organization, it's oftentimes frustrating when I just wished I had someone that understood the backstory, why I started this business, all the things that I've gone through to get here. I wish I had someone that kind of knew me and I could brainstorm with, but oftentimes we have to figure it out. It's lonely at the top, they say. Well, that's where that uh comment's coming from because it feels lonely because we don't have people to brainstorm that understand us. We could lean on our spouse and they're, you know, but they're still not in your shoes, they still don't understand. So the trick is less tune this model to be more in your shoes to give you more relevant uh input. And so here's here's an exercise I did before the show. So if you use chat GPT a lot, over time it learns about you, okay, and then it it's their way, it's chat's way or Gemini's way or Grok's way of starting to bring you content that's cure curated, that's just for you. Our world is moving from a one size fits all answer to a very relevant, specific answer just for you, knowing all the things and your your um proclivities and the way that you you want to go about the world and wants to bring back a unique answer that just hits, right? So we're in that journey, we're in that timeline, we're somewhere starting to enter the early stage of that, and so you still have to work with it. So it's it's trying to learn you, and so I've used it extensively for a couple of years now, and so I thought, you know, that how would one like you get Chat GPT to help you fine-tune it to where it's not so sycophantic, right? So I put in here uh, you have some knowledge about me, and I wanted you to describe me, and and I'm gonna take this prompt. I'm getting so I'm telling it, here's what I'm expecting from you. Here's my intent with this. I'm gonna be prompting in Chat GPT in the settings so that when I'm working with you, I have an expectation that you're you're making sure I'm looking at opportunities, challenge, and scenarios, strategies through a lens that is healthy and aligned with my intent, my perspective, the way I see the world. Now, assuming that you understand that, what would be the prompt? And then I say, well, first of all, list out, I just want to check and see if it really knows me. Show me if you understand that or not. Okay. And so it says, here's what I understand about your lens, the Steve Brown operating system, if you will. And so it goes through and it tells me how I see the world. And this is then it designs a little prompt and it says, when working with you, I should filter through alignment, make sure ideas, strategies, or systems are congruent with your intent, honest, sustainable, empowering for others, spot leverage points, identify when clarity, automation, or story can multiply results without losing soul, challenge lazy thinking, push back gently but firmly against ideas that seem clever, but aren't grounded in reality, ethics, or effectiveness. Preserve clarity, translate complexity, complexity into frameworks that create mental and emotional order, encourage sovereignty, help you and those who serve, build autonomy, financial, intellectually, and spiritually, not dependence, and zoom in and out, stay optimistic but discerning. And so here's gives me the prompt. And then I said, okay, so now this is a prompt I found on X that's telling it to be blunt, but I want you to take into consideration my view and weave in the insights from this brom, this uh uh this blunt prompt because I want it to be more aligned with me, but I kind of like some of the things in it. So it came in, it tells me, oh, it loves that intent. See there, it's being sycophantic. But now, now it fine-tuned it, and so now I have I have a way to take and paste this either at the top of the chat or I'm gonna show you where to put it in your settings. But alignment first, exacting, uh, critique, no theatric, sovereignty, over dependency from story to system, zoom levels, discernment radar, rigor by default. So let's let's say that we're happy with this. So this is where you go. This is this is how we covered the why. This is the how you can get it to think more like you and get it to be more um strategic and aligned with you, a thinking partner. And so here's where we go. So in in your in chat GPD specifically, down here on the bottom, you would click and you are going to click settings, and it's gonna take you to all of these. And so in the personalization area, this is where you would put in that prompt. So let's say that, let's say that um this is what I want it to do. Let's just say I copy this, and for the sake of this, I'm I'm not going to go through it necessarily, but this is where I would paste that. Okay. Right now I have a very generic, superficial. Um, take a forward-thinking view, be innovative, and think outside the box, be empathetic and understanding, and use quick and clever humor when appropriate. So I would take it and pop it in here. But let's let's uh stop for a second and just review these. So, first of all, you can give Chat GPT a personality. So when you pop these down, it's got default, cheerful and adaptive and uh sycophantic, right? And then we got cynic, critical and sarcastic, robot, efficient and blunt, listener, thoughtful and supportive, nerd, exploratory and enthusiastic. So I picked nerd setting. Then we have our custom instructions, and then we have a little bit about me, but then we can put more about you. So this is where we can take um this section here and pop it in to this more about you. Then here's where the memory reference and say memories, this setting will be on, and that's the reason you would turn this on, is that you would want it to learn more about you and be again. Remember where we are on that journey, we're going from a one-size, generalized, superficial answer that we have to uh dig around and spend a waste a lot of time until we find something that might be a little bit uh helpful to not too far away where anytime we ask, we're gonna get curated content that is impactful, relevant, and very helpful because it understands everything about you. And so turning this on helps tune this model to be more uh understanding about you, and then down to do I allow it to do web search, etc. And so that's what those settings are, and so this is a way to tune chat GPT. You know, oftentimes I'm talking about using custom GPTs, okay. Custom GPTs, your strategic partner who turns your knowledge into real results. Well, this is a way to start. This is the very first step that you can do that can help you over time learn about you. A custom GPT is how we can take this model and leapfrog years of understanding you and bring it up to speed, like bringing an employee on. And instead of having a six-month uh onboarding and learning and training and orientation, we can orient this model to be impactful immediately by doing a custom GPT. So that's the difference between tuning this immediately, that over time it really dials in and starts to give you relevant, impactful, helpful information too. Hey, let's speed this up and let's turn this into an um a thinking partner for not just me, the leader of the organization, but for my whole organization, they can start using this custom GPT, and it's like they understand me and they think me and they under we get all the backstory, and it starts to be infused in all of the content you produce, all of the strategy, all of the brainstorming. That's why custom GPTs are so powerful. So when we learn today, we learned how AI tends to agree with you, they're tuned that way. Okay, they're just come out of the box that way, but now we're starting to know how we can put it into your DNA. The hidden difference between blunt and useful feedback, we got that covered in how to tune Chat GPT's personal personalization. So if you're like Steve, uh I love this. I'm learning this, this is something that I've been feeling we need to do. Well, here's what you do: you come in here, set up a time, and we'll create a plan. I'll learn about you, and then we'll move forward on this together and set you up for success to excel in the world. This new world were is uh coming at us. Like going uh 150 miles on a highway. Everything's coming at us so fast now, and you need to be set up to be successful and be able to see around the corner. That's what we do. So, if you want to work with us, schedule time, let's do a custom GPT. But in the meantime, tune, tune your your um model that you use, get in there. It's gonna set you up for success. I'm Steve Brown. This is AI Made Simple. If you like it, if you're getting some tra uh if you're uh I'm getting trouble screaming streaming to Facebook. Who needs Facebook anyway? If you're getting some value out of this, like, subscribe, tell your friends and neighbors, bring bring them all around, and let's set you up for success. See you in the next episode of AI people.