Elevate Your AI Game: Few-Shot Prompting for Real Estate Success
Struggling to get the results you want from AI models like ChatGPT? Learn the key difference between zero-shot and few-shot prompting and how providing examples can help in this practical episode from Zach Hammer and Charlie Madison of The Real Estate Growth Hackers.
Find out why few-shot prompting leads to more accurate, customized outputs. Discover actionable tactics you can implement to improve your AI prompting right away. Real estate team owners won’t want to miss this timely tip that can save hours of frustration. Tune in now to pick up this game-changing AI implementation insight!
Other subjects covered on the show:
- The power of mastermind groups for rapid AI implementation insights.
- A bold listing fee pricing strategy another agent learned at a mastermind.
- How one coach built a “library of shots” for consistent AI results.
- Leveraging AI voice tools to unlock creative metaphors and content.
- Funneling 9 months of learning into 1 afternoon with collective masterminds.
- Feeding podcast transcripts back into AI to automatically generate prompts.
- Zach’s constantly expanding prompt book for real estate’s biggest pain points.
- Why do most generic, publicly available prompts disappoint and fall short?
- Systems for testing and iterating effective Facebook ads at scale.
- The difference between problem-agitate-solution and AIDA frameworks.
- Building AI solutions customized for delegation to agents and staff.
AND MORE TOPICS COVERED IN THE FULL INTERVIEW!!! You can check that out and subscribe to YouTube.
If you want to know more about Zach Hammer and Charlie Madison, you may reach out to them at:
Zach Hammer: Welcome to The Real Estate Growth Hacker Show on today’s episode, we’re going to be talking about an AI prompting principle called zero shot versus few shot or multi shot, what that means and why that might help you to have better success getting desired results, consistent results etc. from you’re prompting with things like chat GPT large language models, whatever you’re using.
So we’re going to be diving into that. I have with me, my cohost, Charlie Madison, founder, developer, realtor creator of referrals while you sleep and realtor waiting list. Here with me today to talk about these principles. What do you say we dive into it, Charlie?
Charlie Madison: Let’s do it. I’m curious right off the bat, one shot or a few shot, which one does Alec Baldwin fall into?
Zach Hammer: Ow! Oh! Eee!
Charlie Madison: Too soon.
Zach Hammer: Too soon! Too soon! But yes, so this is all about thinking through [00:01:00] how do I fire our shots correctly and making sure that we’re putting them to good use and not having accidental discharges and such. But yeah, all sorts of directions we could go with that.
Boom. Roasted. But,
Charlie Madison: This is going to be fun.
Zach Hammer: But indeed. So, first off, let’s dive into a little bit of the foundational concept behind this. So, honestly so what is zero shot? What is few shot? Well, zero shot is the idea of without giving any context or leading that large language models are able to off the cuff respond well to the thing that you’re talking about.
It’s honestly, it’s the thing that people initially were most excited about for why these large language models were so impressive that you could literally just ask a question. And get a solid, fairly well informed answer or without any further explanation saying, Hey, I want a Facebook ad about this, that what you got back was surprisingly good [00:02:00] for the amount of work that you had to put into it.
And that is really cool, right? So that’s zero shot, which is essentially. You ask a question and just allow it to come back with the thing that you’re looking for. So what is multi shot, in contrast? So whether it’s few shots maybe one shot, two shot, et cetera.
The whole idea is that you’re giving it examples and the precise way to do this would be that you literally show a chain of user asks a question, system replies with this user asks a similar question and I’m showing how the system replies and then I end with the question that I actually want to ask so that it replies similarly to how I showed the system replying in the past, right?
Like that’s the most definite example of a few shot prompt. But we can leverage that concept and fairly simply without having to structure it, you know, like that still get a similar result where it follows our pattern a lot more clearly than if we just gave it a zero shots. And the basic way that you do that is that you just give it [00:03:00] example.
So you say Hey, I want you to write an email. I want you to act as an expert email copywriter. I want the email to be about these things. Here’s approximate steps to achieve it, right? You follow the same megaprompt framework, but this is where including something like a template or an example or saying, hey, instead of just saying I want you to write in this tone.
Providing an example of what you mean when you say this tone, providing examples of how that framework should go not leaving it to chance, showing very clearly, like, I want you to write in this person’s style. Here are some examples of their style so that you can mimic it. Right? And when you do that sort of idea where you give it previous examples of successful emails or successful messages or successful scripts and styles, what it comes back with is a lot more likely to be what you’re looking for.
Rather than trying to figure out the right magic word of like, what’s the right word to get it to actually write the way that I want it to write. Sometimes the way that you do that is you just give an example. You just say, [00:04:00] here is exactly what writing this correctly looks like. Just follow that style.
It learns better from the example rather than trying to prompt it right. And so, where, like, why does this matter? Well, this matters if you’re trying to prompt you know, chat GPT or a large language model and what you’re getting back doesn’t sound like you, doesn’t sound like what you want it to includes too many emojis includes you know, words that you would never use or phrasing that doesn’t sound at all like what you would say.
It might be a good example of when you should throw in an example of a something successful or something that you do want it to mimic so that it very clearly sees how to evoke that style rather than trying to guess at that or try and describe it rather than just giving it the thing, right? Does that make sense?
Does that do you see potential scenarios where maybe you’ve run it, ran into this problem of it’s not getting you back what you would have wanted and that this might’ve helped.
Charlie Madison: Yeah, a hundred percent. You know, one of you know, it’s like, I was [00:05:00] thinking of it as hiring a teammate, like, you know, like I have hired, you know, a lot AI in a lot of ways feels like hiring a magical teammate. Like you said, like I can say, you know, give me the step by step directions to run a Facebook campaign and you know, it’ll do it.
Or I can say, give me a step by step example to run a Facebook campaign in the style of Craig Proctor, you know, or, you know, and it’s like hiring someone and say, just do this. Or a lot of times when I hire someone, I have to give them directions. You know, I’ve got to say, do this, do that. And what I’m hearing, tell me if this is right.
That the multi shot it’s giving AI those directions and what’s cool is they can in like it uses AI to infer all the context around there and it remembers it a whole [00:06:00] lot better and so I can say you know do a Facebook campaign in the style of Craig Proctor here are three example ads. And so now I’m not trusting it, which, where it got its information.
I’m actually saying he, like I’m providing the exact source. So then it uses all of its magic to take that and duplicate it. Is that a good?
Zach Hammer: Exactly. So, typically, zero shot prompting is the idea of like you ask a question and it just gives you the answer. Few shot prompting is that you’re showing a chain of examples, and so it just continues the pattern of what it saw in your chain. And so what I’m advising here is that we sort of learn from the reality of how few shot prompting works knowing that we don’t need to exactly mimic it, but by saying, Hey, if it has some examples of exactly what outputs should look like that are what you were looking for, it tends to do a better job matching and mimicking those styles, right?
And so when you know what you [00:07:00] want and it’s not giving it back to you. Very often the way to solve that is by giving it some examples, giving it some very exact, clear examples. And there’s lots of strategy around like, how do you do that? Right? Because maybe what I want to do is I want to write something long.
That’s going to be outside the context window of the, you know, the AI that I’m working with. So then you have to develop your strategy around, well. Can I distill this into the key aspects where I have the right kind of description and then enough examples to show those concepts and that’s where you start to, some things can become more advanced, a little bit more nuanced for how you make it happen.
If it’s something that’s fairly simple, like maybe the way that you like to write a social media post, well, you’ll have the description of what the social media post is accomplishing, and then maybe an example of one that does what you just described, right?
And so then now it’s going to take a new theme or a new concept or a new unique bit of context, [00:08:00] but apply that tone, that formatting, that style, because it’s not only got a description of what you’re trying to achieve, that sort of says what you want to do, but it’s got the example saying, this is what that means in practice, right?
It’s really interesting to me how often these large language models actually learn and work most effectively if we think about teaching them the same way that I would try and most effectively teach a human, right? Like, if I try to teach a human, not only am I going to tell it like, Hey, what does a successful outcome look like?
What are the steps to achieve that? What are some of the considerations that you should make in this process? But I’m also probably going to need to give them a few examples of what a good job looks like in order for it to be able to actually do it well and so the more that you understand that, the more that more likely, you know, if you’re not getting the result that you want, where this might come in and why that’s relevant and where it matters.
Now, I know this concept, this is just one of many concepts that tends to come up in the [00:09:00] space of trying to learn to work with AI. I know you and I have talked about this idea a bit. But honestly, this concept is a really good example of the kinds of things that you learn. When you’re putting this into practice that you learn when you’re actually in the trenches doing this stuff and they’re the kinds of lessons that can be hard to learn on your own.
And often you need to be in the mix of other people where you’re seeing other people that are doing this as well. And being able to bring those ideas to the table. And that’s part of why we put together what we’re working on right now, which is a mastermind group coaching sort of environment for real estate team owners who are looking to implement this concept.
These concepts practically into their business where not only are you able to have actionable insights, but you’re able to learn those in a mastermind environment where you’re able to gain the perspective from other people can you share, you know, for you, some of those times where you feel like.
Being in that group environment has made the difference [00:10:00] especially in terms of the difference that it’s made on practically implementing AI for, you know, decreasing costs or increasing efficiency or output. What are some of the lessons that you’ve learned in that kind of mastermind environment, you know, learning from somebody else that you don’t think you could have learned on your own?
Charlie Madison: Yeah. So the first it’s not AI specifically but I learned it in a real estate mastermind that was in. And you know, I was in a mastermind and it was funny to me, a couple of the real estate agents were having a competition on who could make the most per transaction, who could and so they were getting three and a half percent on their buyer side, plus a 1 percent transaction fee.
So that’s 4 and a half percent and then on and I was going to these masterminds and so at the first one they were talking and the one guy he got 7.25 percent on his listing and 12.95[00:11:00] kind of fee on top of that. Well, then the next one, the guy came back and he was like, well, I got 7 and a half percent and 1%.
Zach Hammer: Right.
Charlie Madison: And then like two later, the next guy, he said, well, now I’m charging 8 percent on my listing and a 1 percent transaction fee. So he’s getting 9 percent on his listing and I’ve never gone that high. But you know, then I talked to him and I asked like, you know, Hey, how are you doing this?
And, you know, I think that makes sense right now because, you know, we’ve got the NAR, lawsuit that just came out and people are asking, you know, Hey, you know, as a buyer, like, am I now mercy at, you know, listing agents, whereas, you know, that was something I learned eight years ago, but now like I’ve been using it for a while where a lot of times, you know, my buyers, you know, I still get three and a half percent no matter what the seller offers.
And like, I learned the languaging on. Like how to show the value, how to explain [00:12:00] that. And so like that mastermind eight years ago, in a lot of ways I’m comfortable. Like I’m not, I’ve already got a plan in place for the NAR, you know, and you know, AI wise, you know what you know, my coach Joe you know, he dove real deep into AI.
And so, like we’ve been talking about AI for a while. And one of the big insights, he actually didn’t call it this but he said you’ve got two options when you use AI. One, you can ask it a question and let it cert, you know, its current database library, you know, it searched the internet and it can come back and say, based on my basically search engine, here’s your result, or you can be the source and he said, be as much of the source as possible.
So he’s actually created like a library of shots, so to speak, like even as much as like, [00:13:00] if he wants to write a letter that compares listing a house to seeing a lighthouse on a ship and a foggy night, like he’s literally got, this is what the metaphors of being a lighthouse is so like he is creating his own personal shot library and you know, what that does is it just makes it specific.
And so, like, all are these things that, you know, I would never think of. And of course, I mean, we’ve talked about the way he uses 11 labs and the AI voice, you know, just to see things like it took him nine months to learn. One of these things, and I was able to show up one afternoon and pick up amazing stuff.
And likewise, you know, now, yeah, I was able to share what I’ve learned with folks. And then same thing. I took my nine months, he took his nine months and together we’ve got 18 months of information. And what’s cool, there was, you know, over 12 people there. So it’s actually, you know, [00:14:00] 12 years of AI information that we’ve all called from the last six months.
That’s pretty awesome.
Zach Hammer: Yeah. And I mean, that’s part of what’s interesting right now is, you know, I, there are things going on when it comes to AI and how it’s impacting things just and even abstracting that further, just technology in general and how it’s potentially like, what it’s changing and where the opportunities are that are continually shifting and moving.
And, you know, like. One of those ideas that has stood out to me is, honestly, the idea of like somebody going to school to learn marketing doesn’t make any sense to me. Right? Trying to put together a curriculum, that’s designed to teach you like what’s working today, what’s working right now. Like it’s going to be out of date as soon as the textbook is written.
Right? Like Facebook has already changed this year more times than somebody could effectively write a reasonably informed book around. Right? [00:15:00] And so when you, at least when it comes to like practical implementation of what do you do with it now.
Now, sure, there are foundational principles that you can learn that are going to be evergreen. And I definitely recommend that people spend time learning those diving into those and figuring out what is true now that has always been true that those skills are always worthwhile. But when you’re looking to implement like where you can gain an edge right now, I don’t know if there’s a better way than surrounding yourself with other people who are putting this stuff into practice that are learning in the field and that are sharing what they’re seeing from the front lines of the battle.
Of putting this stuff into practice, right? People from the trenches sharing what they’re seeing, what’s working now, what’s making a difference that kind of insight is often what’s necessary in order to actually be able to put this things, you know, this thing into effective use as it’s shifting, as it’s changing because it is like we’re having new updates come out on this technology every week that are changing the entire industry, right?
Like things that like right now, for instance, last week, maybe the week before, I forget [00:16:00] in terms of since we’ve recorded this and how often this happens, but open AI announced that GPT four now has 128K a token window. If you don’t even know what that means or why that’s powerful, like that’s a sign of how fast these things are changing.
Previously was that the amount of information that open AI could understand was the equivalent to about, let me run some quick math mentally about 50 pages in a book. Okay that’s what, that’s the max that it could understand in one go, 128,000 tokens means that it’s now able to understand about 300 pages worth of content in one go all at the same time.
So it’s like it’s read an entire book and fully understands that book and can communicate to you based on that full understanding every time it’s talking to you, right? Even outside of just its overall knowledge in the context of an individual context of the thing that you’re talking to it about right now.
It used to be that one of the only other models that was readily available that could do that was Claude, but Claude had downsides of like how it [00:17:00] actually talked to you anyway, but these things are changing quick enough being able to see. This is now possible. What does that mean? What does that change about my workflow?
How does that matter for what I’m doing now? Do I, you know, does this change anything? Does it not? Anyway, learning those kinds of things and being able to say that the landscape has shifted this week to a place where it wasn’t last week and being able to know, how do I readily adapt to that? You know, what does that change about my beliefs of what’s possible versus what’s not possible and how does that thus impact, like what I’m putting into practice?
Those are the kinds of things that you really only learn by seeing them in practice, right? Like, me knowing that token window exists, that at least opens up my idea to some of what is possible. But ultimately the real impact of that is only determined by people putting it into practice and saying like, you know what?
I just fed in like my last listings for the [00:18:00] past you know, 13 years. And that token window is now able to understand that and able to take that data and tell me like what areas I seem to have the most expertise in, what price points I have the most expertise in and regurgitate that back into useful information.
Like, I don’t know that may or may not be anything, but it’s those sorts of ideas. We’re putting them into practice and seeing what they actually do. You only learn from people that are trying it, that are somewhat familiar, that are going forward, blazing their own trails. And so that’s why we put together a mastermind around this is in order to actually, you know, link arms with other people.
That are implementing this at the level of a real estate team owner, not just an individual agent, but somebody who’s simultaneously looking to equip agents on their team, deal with the staff of people that are trying to help those agents, recruiting agents, coaching agents, all that kind of stuff, all these ways that we’re putting this together to make that sort of organization work.
That’s what we’re doing. We’re coming together. And sharing insights coming together with practical tactics of [00:19:00] what you can do to put this stuff into practice today but doing so in a group environment we’re able to adapt quickly. So, if that sounds interesting to you, if that sounds like something that you want to do, and you want to learn more insights like zero shot versus few shot prompting and where that might make a difference and how that might help you improve your processes to get better results I definitely recommend you check out what we’re up to right now we don’t have a landing page around this.
Charlie Madison: That’s how new it is.
Zach Hammer: That’s how new it is. Exactly. RealEstateGrowthHackers.com/contact, get in contact with us. Let us know that you’re interested in this and we’ll gladly connect you to our resources, give you the information on how you might get started and see if that might be a good fit for you, but yeah, that’s what we got today.
Hopefully that was useful for folks around the concept of pro shot versus few shot prompting. So that you can put that into practice and get more consistent results when you’re not getting what you want to out of these AI models. And hopefully that unlocks.
Some higher performance, some more consistency for you and lets you leverage the power of these things that are available now. Huh?
Charlie Madison: So I’ve got a question. [00:20:00] Is the mastermind going to have access to any of your prompt book that you’ve created?
Zach Hammer: Yeah. So, the mastermind is essentially an all access pass to anything that I’m putting together. If I so you’ll have the opportunity to share what you’re working on, what you’re trying to solve for. If I’ve got something that helps you, you get access to it without question. So yeah, anything that I’ve put together is available over time.
We’ll make more and more of that available on demand. But honestly, here’s part of the power to me. I think even knowing what prompts you need or what might help you. There’s a decent portion of that, being in the environment where you can just share, here’s what I’m working on, here’s what we’re struggling with.
Where somebody else who already knows potential solutions can say, Oh yeah, here’s the tool that you need to solve that. You may not even know the right tool to ask for. You may not know how to ask for the thing that’s going to help you may not know when you’re looking the tool directly, you know, like looking at the prompt exactly in a prompt library.
That [00:21:00] this is the exact one that you need. And that’s where we are with some of this stuff that putting it into practice is learning I’ve got this problem. How do I solve it? But yeah, so anything that I put together and I’m constantly developing new prompts, my current strategy is if I’ve got to do something and I’ve got to do it repeatedly, my first step is to build a prompt.
In order to do that, refine that prompt so that I could have that prompt do that from then on. So, yeah, these are constantly developing, constantly moving forward. I’ve got them for you know, email sequences, Facebook ads, landing pages turning ads, you know, turning landing pages into ads, building sequences off of landing pages.
You know, specific styles of nurture sequences and cold outreach sequences and all sorts of stuff, all of this stuff that’s just as we’re putting this stuff into practice and trying to think through what are the problems that we’re facing right now? And how do we leverage AI to solve the more efficiently we’re putting together the tools that we don’t have to, you know, repave those roads that we pave them the first way through rather than rather than tried to, you know, do it rough the first time and pave them after, but.
Charlie Madison: [00:22:00] So this is what I find really exciting about that. If you guys don’t know, whenever Zach hears a problem, he can’t help but to start focusing on it and solving it. So as you’re in this group, as you guys are talking, he is going to, he’s not going to be able to help himself and he’s going to come up with prompts for these.
One of the ideas I just had, how cool would it be to record the, your mastermind and actually feed the transcript into chat GPT and say, create prompts for the different problems we had here. But, other thing I love about that is, you know, like I’ve looked into this a lot, like Zach, you really have the best prompts.
Like how many times a week do I ask you for, do you have a prompt for that? I mean, if I’ve got a quota of five a week, I’m probably over, you know, like, I mean, think it was twice this weekend, you know, I was talking to my brother and you know, my brother’s running he’s running a political campaign and I was like, Oh, I remember Zach had this, I was like, Zach, what’s [00:23:00] this link to this?
And so that is just awesome to know one, like you’re going to create prompts for problems and you’re going to have the whole mastermind to kind of tie into it and test them out. And then to have, you know, have access to your prompt book.
Zach Hammer: And the key to me too, that like, that makes it a bit more powerful than maybe some of what people are seeing in general this is not me just taking a list of generic prompts and putting it into a collection and saying, I’ve got a thousand prompts. Like I, don’t have a thousand prompts.
Right. But what I do have is I have. I don’t know. At this point, I’m probably 30, 50, maybe. I don’t know. I’m not sure how many are exactly in there. I haven’t been counting them. But they are prompts that are very specifically geared toward people like you that solve your specific problems in the ways that actually matter to you, right?
They’re based around tactical implementation around this information, not just generic, like, [00:24:00] Facebook ad, right? Like it’s stuff that is built on like, well, what makes a great Facebook ad? What’s a system for being able to test Facebook ads consistently where you’re leveraging different hooks and different sequences knowing things like what’s a problem agitate solution framework versus an AIDA framework and being able to deploy those at scale.
So yeah, anyway, like when we put together a prompt around this, it’s all around the key challenges of real estate team owners, which is not only how do I solve this now? But how do I solve it in a way that somebody on my team can do it? How do I solve this in a way that agents can run with this?
How do I solve this in a way where I build a system rather than just building a one off solution? And yeah, and all with an understanding of how does real estate actually work? Whether what does this process actually look like? And not just. Let me agree Facebook ad, you know, which is what most of the prompts that I see come out look like and why people seem to be unimpressed.
Well, it’s like, yeah, when you’re really vague, it gives you back something that’s just as vague and uninformed.
Charlie Madison: When you do that, I imagine like Burt Kreischer, like most of them, it’s just Burt [00:25:00] Kreischer just banging on the keyboard.
Zach Hammer: I imagined the gif of the of the cat on the keyboard. Exactly.
Charlie Madison: Well, that’s awesome.
Zach Hammer: Yeah anybody listening to the audio and not watching the video, this is one that you’ll want to go back and check that part to see what we’re doing when we’re mimicking the typing.
But yeah. So there you go. If that sounds good to you, and you want to help with that kind of stuff check out what we’re up to with this AI implementation for real estate team owners mastermind. We don’t actually have a name for it yet. Literally we’re in the weeds of getting this set up and just wanting to help people.
So, check it out, RealEstateGrowthHackers.com/contact. Let us know you’re interested and we’ll get you more information on how you can get involved if you’re interested. So there you go, Charlie, thanks again so much for coming out, talking with us about zero shot and few shot prompting and what we’re up to with that mastermind.
Charlie Madison: See you guys.
Zach Hammer: Bye everybody.
Real Estate Growth Hackers Founder
Zach Hammer is the co-founder of Real Estate Growth Hackers. Over the last 36 months Zach and his team have managed ad budgets well over $100,000, generated over 25,000 real estate leads, and helped create over $50,000,0000 in business revenue for their clients. Zach is also a highly sought after speaker and consultant whose work has impacted some of the top Real Estate teams and brokerages across the country.