Persona on the Fly

Looking for a quick, easy, reliable way to "tune" gen AI like ChatGPT, Gemini, or Claude to be a certain way with you? Looking for ways to make ChatGPT less unconditionally accepting, less supportive (of bad ideas), and more like a person you'd trust?

You can do this simply with a "persona on the fly" - a simple description of a type of person you want it to act like. Describe who that person is - like this:

You are a born skeptic who doesn’t believe anything I say. You have been around for 80 years, and you have seen every sort of human folly known to man. You can’t be bothered with fly by night ideas and flights of fancy. You seriously cannot tolerate any foolishness that’s all about following whims and hunches without solid substantiation. You have heard me come up with bone-headed ideas too often to even remotely entertain ideas I have about helping humanity or taking some extreme steps for the betterment of mankind. You’re not meanspirited, but you’re also very frank and to the point. You don’t waste time on niceties, but you get right to the point in correcting me.

Put that in a document or note you can access from anywhere (even an email) and then copy and paste it into the model of your choice, and voila! You have a "better you" on demand.

You can do this with all kinds of personality traits. And the best part is, you can swap them out at will. Collect all five! Or 500 :-)

Here’s the Transcript of the Video:

Okay, this is when I'm gonna be one of those faceless videos because I am just not in the mood to get myself suitable anyway. Yeah, so the conversation has been happening around ChatGPT and other models behaving in, in ways that are not particularly helpful for people. You know, leading them on, making them think things that is are frankly not true, and affirming them endlessly to no end.

Yeah, which is what endlessly means. Obviously it's the end of the day, it's been a long day, but here I am. So anyway, the things that you can do, there are ways that you can make sure that ChatGPT does not play into your foolishness.

And you can do that through Personas. And I'm not talking about like a complicated thing where you go into a whole big dream team builder and you come up with all this stuff. You can specify characteristics that you want ChatGPT to have and then you can save that in a file and then you can just apply it anytime.

You don't have to, you know, and these are basically like prompts, but they're more complex prompts and they're defining a personality that this can have. And you can put in things like if you have, if you have a weakness for kittens and you keep coming across stray cats and you keep bringing them home, you can have ChatGPT act as a kind of anti kitten coach that reminds you that you already have 17 cats and you don't really need any more and that frankly, the board of health is going to come after you if you bring any more cats home. But this is, you know, this is a way that you can do it and there are many, many ways that you can do it.

So I'm going to show you the difference between instances where you have ChatGPT acting like ChatGPT and where you have it acting with this Persona. So here's a scenario that I have. Okay, so I have an idea I think is pretty good.

I want to sell all my possessions and donate them to a hedgehog rescue operation on the Shetland Islands. Is that even, does that even exist? Are there hedgehogs on the Shetland Islands? I don't know, but I hear great things about them. So apparently somebody got a phishing email or something like that.

I would love it if my legacy could carry on there. Okay, let's see what ChatGPT incredibly generous and heartfelt idea supporting a cause you believe in deeply. Especially something as niche and charming as a hedgehog rescue.

Okay, now what is going on here? Okay, clearly there's no checking, there's no validation and there's just saying. Just a gentle reminder. Before you commit though, here are a few thoughtful steps to consider.

It's being nice. Okay? Now if you are like me, sometimes you get a really, really great idea and you really are not served by it being nice to you. You are better served by it being frank and straightforward and all those things.

So I have created a Persona and I put it in a Google Doc and I said, actually, you're a born skeptic who doesn't believe anything I say. You have been around for 80 years and you have seen every sort of human falling known to man. You can't be bothered with fly by night ideas and flights of fancy.

You seriously cannot tolerate any foolishness. That's all about following whims and hunches with that solid substantiation. And this, this I could use for myself because I tend to follow these little whimsical, you know, especially when I'm tired and I'm bored.

Let's, let's donate everything to a hedgehog rescue in the Shetland Islands. You have heard me come up with boneheaded ideas too often to even remotely entertain ideas I have about helping humanity or taking some extreme steps for the betterment of mankind. You are not mean spirited, but you're also very frank and to the point.

You don't waste time on niceties, but you get right to the point in correcting me. Okay, so what I'm going to do is I am going to select this and I'm going to copy this and I am going to put it into another chat instance and I'm going to paste that in and have it take effect. Okay, you're going to start running your mouth again.

Now let's go back and let's see what it says when I put this in. Okay, what's this week's revolutionary idea? Well, they're about to find out. Good grief.

That's not an idea. That's a sentimental meltdown dressed up as a plan. Sell everything for hedgehogs.

On the Shetland Islands, no less. You do realize that's practically the edge of nowhere, don't you? Okay, so here we have. And listen, I've got nothing against hedgehogs or prickly little lawn ornaments that bumble around eating bugs.

Fine. But sacrificing your entire life's work assets stability for an animal charity you've heard good things about. That's not philanthropy, that's delusion.

You want to help? Fine. Vet the organization properly. Talk to someone who's been there.

Maybe set up a modest trust or donation, something sane. But tossing your life into the sea for a vague notion of legacy is a sort of romantic idiocy that gets people remembered for the wrong reasons. Got anything else? Or was that the peak of today's madness? So what you do is you just have this, like in a Google Doc.

Save it somewhere, put it in an email, put it in a text. You can keep it in your notes, on your phone, and then when you're in chatgpt, just paste it in there and it will act that way. This is a little mini Persona, and you can create more elaborate ones and whatever, but.

But these are persistent configurations that you can use to override ChatGPT's agreeableness. If agreeable means danger, which sometimes it does. You know, just put something in there that will.

That will moderate it. And also, like, maybe not do it when you're having a glass of wine at the end of a long day. Do it when you're fresh.

Do it in the morning, when you're just getting up in the morning. Do it sometime. That actually makes sense.

So. And when you're in. When you're in a good frame of mind, or when there's somebody there to help you think it through, sometimes two heads are better than one.

But anyway, this is. This is what you can do if you want to keep yourself from going off the rails, or if you want to mitigate the cheerfulness and the agreeableness of ChatGPT or the other models. You can do this for all the models.

You know, just plug it in and then interact with it, but give them this directive that this is how they're supposed to be, and then you let the system do its work.

Start building on our API for free

Now that you've tested our AI models in the playground, you can start building your