#197 - AI for CPG 101 with Chris Szigeti
Chris Szigeti
For those foundational models with those broad models, like, the most common is again, ChatGPT Quad Gemini, which is Google's, and there's Grok, which is X's. You know, there's a handful of them out there. Then there's Perplexity as well, which gets brought up a bunch. It's pretty foundational, but it's geared towards search, so it's a little bit different. But these are those foundational, those broad models. My advice to people starting to experiment or use these is like, don't get analysis paralysis. Pick one, use it. They are all so similar, and when they do release updates, they converge so fast that if Grok pushes out a new feature, ChatGPT is going to have it within a month or two.
00:53
Daniel Scharff
Welcome, everybody. Today is the day that we actually do an AI for CPG101. We've done some really cool content featuring brands that are being very clever and figuring out how to use it well in different ways. But we've never just started from the beginning to catch everybody up to speed so we can all be experts. So on today's episode, we've got Chris Zaghetti. Chris comes from that world. He's a tech consultant, product strategy manager, but now he is on the CPG founder journey, launching his own brand. So he really knows how to relate all of the stuff that's out there in a way that we will all understand and start to understand how to use. So on today's podcast, he takes us through what. What is AI? What tools are out there?
01:32
Daniel Scharff
How should I actually use them in a really effective and efficient way? He talks about a lot of use cases for cpg, some of the pros and cons, and what are just some of the tools out there that we should know about. This episode is a total banger. I hope you love it. Let's go. All right, welcome all you listeners, and welcome to our robot overlords, who are probably also listening. All of the bots out there, what's up? Be nice to us when you take over. I'm very excited about today's episode because we have got Chris Sighetti on here. And not only is Chris a burgeoning CPG founder, but also he is a super tech pro. He is a product strategy consultant to tech companies and knows all about the AI stuff.
02:19
Daniel Scharff
So what we've got here is the perfectly positioned person to explain this scary world of AI to all of us. So, first of all, Chris, welcome. Thank you for joining us.
02:29
Chris Szigeti
Oh, thanks for having me, Daniel. It's a Pleasure to be here or this podcast has been just super helpful in kind of my CPG founder journey.
02:35
Daniel Scharff
Okay, Chris, are you real or are you a bot?
02:38
Chris Szigeti
A little bit of both. I am mainly real, but it feels like I am going more towards bot every day with how much AI I use, to be completely honest, so.
02:46
Daniel Scharff
All right, sounds like something a bot would say, but I'm going to trust that you're mostly real, so let's just jump right in. First of all, can you tell people a little bit better than I did about your background?
02:55
Chris Szigeti
Yeah, absolutely. So I've spent the better part of the last decade helping grow, launch, build tech startups. So not CPG very much in kind of the tech world. Building real estate, tech climate, technology, finance, apps a little bit all over the place. And my focus has been mostly in like product strategy and development. Really help figure out, like, what are we building, you know, why are we building it? And like, how do we actually get it in the hands of people. In addition to that, as you mentioned, I'm pretty early on in kind of my CPG founder journey as well. Prelaunch with a brand in the honey space, which has been new and exciting and great to kind of take a step from this world of software and get my hands dirty with some physical products, which is. Which has always been fun.
03:37
Daniel Scharff
And before we just get into all of the stuff we want to learn from you, can you just tell us a little bit more about what do you hope this honey product is going to be? What's going on with it?
03:46
Chris Szigeti
Yeah, so it's called froth and Honey. We're prelaunch. You can go to the website, keep you posted on kind of all things update. But you know, in my software background, I've really looked for opportunities and that's kind of what drew me to the CPG space. You know, I saw a category, honey, which has had very little innovation, to be honest, and kind of the last forever, to be honest. You know, all the brands look the same, all the product, it's honey, right? Like, how much innovation can there really be? And so we're building kind of a unique brand with a bit of a twist around honey. More to come later. But you know, my ambition for that brand is to really grow it. Like I've taken the same approach with tech startups in the past.
04:22
Chris Szigeti
So you don't bring in really early user feedback and research, which is kind of the phase we're in. You know, hopefully grow it. Granted, it's a much new world to me, so it's been great to have resources like this podcast and the startup CPG community to learn all the things that you don't learn, launching, you know, software. Software companies. So it's been an exciting journey.
04:44
Daniel Scharff
I like it. And I also predict for you a wonderful journey of lots of honey and bee puns as you progress.
04:53
Chris Szigeti
There are a lot of them.
04:54
Daniel Scharff
Okay. Our friends at Melody, I offer him a lot of unsolicited puns that I would like for him to use in his marketing. I think the last one where he just kind of shook his head at me was I wanted him to put. I was like, hey, you have an office, like, in your office. What if you had that yellow sign like they have in TED Lasso that says believe? But it was like, believe? He's like, I don't know about that.
05:14
Chris Szigeti
But there is a lot. There's no shortage of bee and honey and buzz puns. You'll hear them all.
05:20
Daniel Scharff
I just don't. I mean, I guess AI is going to do pun better than me and take away, like, the one thing I feel like I still just crush it at.
05:28
Chris Szigeti
Rhymes are one thing it is inherently good at. So, yes, it is going to be the pun overlord.
05:33
Daniel Scharff
Just you have to curate that. Like, pick from all the ones it gives you. That's going to be my new specialty. All right. Okay, so let's get to the good stuff. Chris, I was really excited to bring you on because of all the people that I've talked to, I think you have the best knowledge overall of AI in a way where you can explain it to everybody. So I hope everybody listening is excited to buckle up and go on this journey with us, because we're starting from zero, and by the end of this, you too will know a lot about AI. So let's go. First of all, basic question. What is AI?
06:05
Chris Szigeti
Yeah, it's a big question, right? You can totally nerd out in this space. And I'm going to try not to go too deep because I don't think that's the most helpful at a high level, like when we're talking about AI, especially as it relates to the tools that exist right now and the tools that are relevant to the CPG community. You're really talking about what people refer to as an LLM, Right? You'll hear this phrase a lot. An LLM is just a large language model. Again, nerding out a bit. But in the simplest form, you talk to this thing, you input text, and it outputs text back to you. Right. It's kind of the simplest aspects of the ChatGpts of the world.
06:41
Chris Szigeti
The perplexity, you know, these tools that you've heard of, they're all essentially kind of doing the same thing, which is I put text into it. You know, it is trained on text, which really means the Internet. So all the stuff that's been put on the Internet, it's trained on that text. It scrapes websites, PDF, videos, et cetera, and it uses that to then essentially map what you're asking it to do and then output answer. So text goes in, text goes out. A whole bunch of fancy stuff happens behind the scenes.
07:11
Daniel Scharff
Okay, and that's what an LLM is, or what does LLM mean?
07:14
Chris Szigeti
Yeah, it means large language model. So it's essentially just more or less the type of AI that a chatgpt, again, a quad, any of these tools are, right? And they can do a whole bunch of different stuff, and we can get into that a little bit. The way I like to think about it, and I think the way that clicks to most people is you have what's called like a foundational model. And this is these LLMs, you know, these are tools that are good at a whole bunch of things, but in their essence, you put some text in and it spits some text out. That text might be code. It could even be like getting a little away from text, but images, video, et cetera. But in its essence, you're asking it to do something.
07:51
Chris Szigeti
It's going off and using its knowledge that it's been trained on, you know, through the Internet and other sources, and then it's giving you answer. These foundational models, again, this is like the ChatGPT of the world, the quads, the perplexities tools. A lot of people work here. This is where most people are interacting with AI today, right? You have a ChatGPT subscription, you go, you ask it questions, you interact with it. The reason that I kind of refer to them, and a lot of the industry does, as foundational models, is because all of the other tools, for the most part, are built on top of these. So if you're using, you know, a special tool that goes and creates images, right, it's more likely than not using one of these base models and then is tweaked to kind of give you that output there.
08:34
Chris Szigeti
So for 90, you know, 99% of people, like, what really matters is, you know, you have an LLM, it's foundational models. I go ask it questions and it returns stuff to me.
08:44
Daniel Scharff
So, I mean, this is a step beyond cpg, but those Things that are built then off foundational models. How does that work from a business perspective? I'm like, oh, I want to create something that can spit out awesome images. Do I. Can I just somehow plug that into ChatGPT? Or I have to, like, have a call with them and talk about it, or I pay them to do that. How does that work?
09:04
Chris Szigeti
Yeah, I mean, they're paying them. So when they're hitting them, they're. They're essentially paying a fee per. Per use more or. And again, we can get really into the weeds on this, but the way a lot of these companies start out is these foundational models have gotten so good. You know, 10 years ago, like, were building products, making our own models, and they were a fraction of as good as these new now foundational ones. So rather than recreate the wheel, most companies will use these and then they'll add on top or they'll tweak. So when you start to see, like, what I'll call a verticalized product. Right. So it's a AI product that does something in a specific vertical. This could be legal contracts, for instance, it's using a base foundational model, a ChatGPT perplexity, et cetera. Claude.
09:44
Chris Szigeti
And then it's adding additional context, tweaking, you know, a little bit of stuff on top of it to really get that specific outcome that you want.
09:53
Daniel Scharff
Okay. I thought I had a brilliant AI idea the other day, and then I realized this just is what it does. It exists. Where I was like, oh. Because I spent so much time and money and energy learning other languages. I learned Spanish, I learned Portuguese. The best way for me to do it was just to get a tutor and then just talk to them. You know, once I. Once you have the basics, then like, okay, let's talk. You correct me also while we're talking, but we're like friends and we're just gonna talk, and then I'll start processing faster. I was like, what if you could configure it to do that and you could build it? They're like, yeah, you can just do that, though. You just tell the bot to. You can call it, like, on your phone and just do that.
10:27
Daniel Scharff
That's cool. That's a good use of it. Like, oh, okay, that's not a. That's not a business idea. So that, I mean, but like, that's just basically, could. You can configure ChatGPT to do that for you to just be a. Like a human that talks to you?
10:39
Chris Szigeti
Yeah, totally. And this is, you know, in the technology space especially, this is why it's really exciting time. It's also really challenging for companies because you can build a company and you can build products faster than ever. But you do have this risk of like, how do I differentiate, right? Like ChatGPT and all these other models, they're progressing so fast that like, if you don't have a differentiator, you could create a company in six months from now. They can release something and hey, they do exactly what your company does and you know, they already have access to a billion users or however many it is. So differentiation, really unique value props, just like in cpg, it's super important in the tech space too.
11:15
Daniel Scharff
Okay, so tell me a little bit more about the kind of models that are out there and tools because, yeah, we know Chat GPT, we've heard about it. It's all over the place. My team already is using, I think what you would call maybe another like broad model, Claude. So we use that on the marketing team here. What, like, what else? What other stuff is out there?
11:34
Chris Szigeti
Yeah, so for those foundational models with those broad models, like the most common is again ChatGPT, Claude Gemini, which is Google's, there's Croc, which is X's. You know, there's a handful of them out there. Then there's perplexity as well, which gets brought up a bunch. It's pretty foundational, but it's geared towards search. So it's a little bit different. But these are those foundational, those broad models. My advice to people starting to experiment or use these is like, don't get analysis paralysis. Pick one, use it. They are all so similar and when they do release updates, they converge so fast that if GROK pushes out a new feature, ChatGPT is going to have it within a month or two. Right. So unless you are deep in the weeds of tech and building tech products, like just pick a model, get good with it, use it.
12:22
Chris Szigeti
Like that's my number one advice to people. Like, don't try to find the best because they're all pretty similar. So that's kind of the first. And then in terms of types of models within each of those foundational tools, and I think this is important to get the most out of them. They'll have a couple different sub models. So if you use ChatGPT, there's really three types. There's a general model, which is what most people use. You know, this is great for just asking IT questions, you know, bouncing ideas, you know, brainstorming, that sort of thing. And they have terrible names in like ChatGPT. It's like these are their four zero models that they're 04. They've said they have terrible names, they're going to change them. But those are their general purpose models. They then have what's called like a set of reasoning models.
13:04
Chris Szigeti
And these are essentially a step above those where instead of almost instantly spitting out answer, they're going to sit and think about that, they're going to draw connections and parallels. You know, it could take 10 seconds, it can take like a minute and a half sometimes to get a response and it will spit back out much more thorough like thought process there. So this is really good if you have more complex things or you're trying to prompt the model to get something that's not super straightforward. Right. So an example of when to use a reasoning model. It might be like, hey, my business right now doesn't. I'm a generalize here a little bit. But like, hey, I have a honey business. You know, we do 500k in sales a year. I really want to get to a million sales next year.
13:46
Chris Szigeti
And like these are my three go to market channels. Like how might I expand to a new go to market channel to get there? Right. Again, super general and broad. But what it's going to do is instead of just saying, oh, I know he cares about go to markets. So here's a list of five. He already uses three. So here's the other two. It's gonna really think about what you're trying to do, what you're trying to accomplish, and spit out a much more robust answer. So these are good for like complex tasks, I would say.
14:11
Daniel Scharff
Okay, so just again on like, where is it getting this advice to give you? Let's say in that first model, the general one, where it's gonna spit something out, basically it's looking around all of its sources, it's looking around the web and any research materials that it has access to. And it's looking for instances generally where that text exists and people have asked those kind of questions and spitting back things that people have said or recommended in those kind of times. And maybe that deeper reasoning model is looking a lot deeper, more carefully throwing the things together, like, tell me, I'm lost.
14:46
Chris Szigeti
Yeah, it is complex, to be honest. What's happening behind the scenes. And one of the cool things about those reasoning models is actually in the last month or so they found out that people want to know what's going on behind the scenes. So it will take you through its whole Thought process. Right. So for the general model, I mean, you're right. Essentially what it does is it's trying to match patterns and spit out like what's the next word? More or less. So if you ask it a question, it's going to say, okay, this is what I think. You want the intent and it's going to match based off that, you know, maybe search the Internet, maybe just use, you know, what it knows. Kind of depends on what you're asking it. For the reasoning models, it might go through like a 10 step process.
15:22
Chris Szigeti
Like, oh, Chris's business does 500k a year. It's in the honey space. All right, let me learn about honey then. Okay, these are the three go to market models. Let me now do some research on other go to market channels and then let me compare the delta between the two. Like it'll go step by step, which is. So it's not necessarily that it's doing more research on your question, it's that it's actually thinking about it a bit more like a human would. So, you know, if I ask you a question, you don't just think of one thing and return it to me. Like you go through a thought process and that's kind of what the reasoning models do. So if you're just asking questions like, it's overkill. Right.
15:55
Chris Szigeti
But if you're asking kind of complex things or having it brainstorm or go back and forth with you, I prefer the reasoning models, but they're a lot slower.
16:03
Daniel Scharff
Okay, I just have to ask this question because it's top of mind and you don't have a horse in this race. So from us, I'm just thinking about running all these queries and like how much energy that uses and all these different models that are out there and how much is using up, like what do you think? Like there was that recent kind of meme, whatever, where people were posting what their toy figure would look like. Like, this is me as a kid's toy and it would have all the different things. And I remember there was a little bit of backlash because everyone's like, do you know how much energy all of that stupid stuff just used? So like, yeah, these. When it goes off and thinks it's using energy from somewhere. Right. You could add all of that up.
16:38
Daniel Scharff
So on the pro side, it's leading to efficient thought and resources and it's great and stuff we never imagined. And I use it and it's helpful. And on the downside, yeah, like, you know, might be more energy efficient for your brain Just to think about that. A little bit slower and less. Well, I guess. What do you think?
16:56
Chris Szigeti
I mean, no, that's a good way of putting it. They use a ton of energy, right? Like, and so these general purpose models, they're faster, they use less compute is the term essentially. Like, they just use less energy to get the answer. So these reasoning models, like, yes, they take longer, they take more processing power. This is why the data center industry and like the processor industry is exploding right now. Because you need a lot of data centers, you need a lot of processing chips, right? Like, it is just going exponentially through the roof.
17:24
Chris Szigeti
The interesting thing is there's kind of two, you know, it's also advancing so exponentially that like both the hardware needed to use all this AI magic as well as the way it's set up, the structuring or the software side of it, like, both are getting so efficient at the cost. And the energy use of these models, it's plummeting every month, it seems. So, yes, it uses a ton of energy. It's not going to go. It's going to keep using a ton of energy. It'll get a lot more efficient, both from a hardware and a software perspective. But, you know, that's why there is a huge industry around data centers efficiency, you know.
17:55
Daniel Scharff
Okay, so that's great to hear that it's getting a lot better. So like you as a human, how do you reconcile that? You're like, yeah, it's just, I mean, it's just this is the way we're going and fortunately it seems to be getting a lot better. Or is that how you kind of feel about it?
18:07
Chris Szigeti
Personally, I think, personally, like, I'm really excited about it. I mean, yes, it uses a ton of energy, right? But what. It's also allowing us to do both from like a business and a personal standpoint, like the innovation, the rate of innovation, even every year, it's going so fast that like, we're creating problems. We're solving the problems really fast too, right? Which is exciting. And you know, that is part of kind of the innovation loop or curve. Like when you launch a new thing, like, there's always challenges, right? You know, Tesla's the prime example. Like, they launched their initial cars. You know, there was all this science that said, hey, like, these are not really better for the environment. Right? Which was totally true because of how you get the minerals, the, you know, the mining, the metals needed, the, all of that.
18:49
Chris Szigeti
But over time, as you innovate, you know, it gets more Efficient. You know, you invent new ways to extract the materials cleaner and better. And so, like, that's kind of the curve you're seeing here with AI. Like, first models were in theory, super inefficient. Like, if you look at the models two, three years ago, like, we're just not even on the same level of both usefulness and efficiency. So it's changing really fast. I'm really optimistic about it. We're going to create a ton of new problems. I think we're going to solve the problems too. And it's an exciting time because, you know, you can just build, build so fast.
19:20
Daniel Scharff
Okay, very fair. I. I was just thinking about, you know, like, when, like some kids there, they'll have like a parent who's like, hey, why'd you leave the lights on in that room? Like, you know how much energy it uses? I feel like we need that kind of strict parent for just everyone using AI. Like, hey, why'd you search that dumb thing? That was a waste of energy. It was dumb. Like, you didn't need to know that. That was stupid. Don't do that.
19:44
Chris Szigeti
Yeah, and I, you know, I do not have any data on this, but if I had to guess, it's probably one of those things where the average consumer, you and I hitting it is such a fraction of the usage compared to like the business use cases and the compute and the processing that's going on. Again, I don't have any stats on that, but that would be my guess. Like, you know, I built products that they hit, you know, this into the query, right? They hit the AI way more than I ever would in like a year. So I think in terms of exponential order, like, consumers are probably on the small side.
20:14
Daniel Scharff
That is interesting, I guess. I mean, they say that also similar to things like water usage, right? Where there's just so much that goes into ag and all of that. Okay, great. Okay, so let's get back to actually using the models and let's get a little bit more detailed because I want everyone out there to know as much as, you know, so they can be efficient when they actually use the models and use less energy.
20:36
Chris Szigeti
Of course.
20:38
Daniel Scharff
Tell me, what tips do you have for people? You're in there, you get into ChatGPT, Claude, you know, perplexity, whatever you're using. How should you get in there and actually start using it?
20:47
Chris Szigeti
Yeah, absolutely. I think probably the biggest mistake I see people make is they treat it like Google, right? You have a question, you just input something super simple into it. It'll give you a great answer, right? Like, you know, it'll search all the Internet. I'll get you answer. But you're really just treating it like a Q and A. That's great. You can get value from it. But in order to really maximize the value you get from these tools, you have to treat it more like you're having a conversation with, like a coworker is kind of the way I put it, where you're not just going to walk up to a coworker and say you want to learn about, like, hey, you know, what are the shelf sizes at Whole Foods, right? This is something I asked the other day.
21:24
Chris Szigeti
I'm not just going to say, hey, give me the shelf sizes in Whole Foods, right? We're going to have a conversation about that, right? It's going to say, all right, well, in what space, you know, what region, you're giving it all this context. When you talk to somebody, you don't give Google all that context. And this is kind of the big difference between a lot of these AI models and tools and just typing something into like a search bar, for instance. And so give it a lot of context, right? For my, you know, CPG brand, for instance, I have put in PowerPoint presentations, I've put in reports, I've told it all sorts of things about my business model, you know, the industries I'm going into. So when I prompt it, you know, it has all that context in memory because I've done it.
22:03
Chris Szigeti
But if I'm going to do that for the first time, like, I'm just going to say, hey, I sell honey. I want to get in a Whole Foods. Like, what's the size of the shelf? I'm going to be like, hey, you know, my packaging is this dimensions. You know, I produce this much in my first run. This is the type of packaging I use. You know, I'm looking at going into a retailer in this region. Like, you're going to give it a lot of context. So the more the merrier. Honestly, like you know, you can give it one sentence, you can give it five pages. You know, the more context, the better the response. So, like, that's kind of the first tip and strategy is more context, the merrier. That's always kind of a great approach.
22:36
Chris Szigeti
The challenge that I think a lot of people have is, you know, it's a new way of thinking and interacting with kind of chat interfaces or interacting with like, you know, what typically has been like a search like a Google. And so A lot of the times I'll hear people get, all right, well, how do I actually go about doing that? And the tip that I give a lot of people is ask your question and then say, hey, what clarification or what context do you need to get a better output and what the models will do? So if I go into ChatGPT and I said that same thing, it would then ask me, you know, 5 to 8ish questions. All right, well, tell me about the size of your packaging, right? Tell me about the category of your product.
23:12
Chris Szigeti
You know, it would ask you a bunch of questions to clarify and then it would say, okay, when you hit Enter and it will take both your original prompt, all the clarifications you did, and it will spit out answer that's going to be so much better than if you just said, like, what's the shelf size of Whole Foods for my company? Right. Some more context. You can ask it to ask you what it needs to get a better output or a better response is really important. I mean, kind of the last tip I'll give people too is if you have like a base prompt. The other thing I do a lot is I'll say like, hey, this is my prompt. Enhance it. Right? Super simple. And what it will do is it will take.
23:48
Chris Szigeti
If your prompt is super simple, like, what's the shelf size at Whole Foods? You enhance it will make like a whole page of details and context and examples. So again, these are kind of quick hacks to really improve your prompting where, like, you don't need to be a master in it. You can ask the AI the best way to do things. And I find that's like the quickest, the dirtiest, easiest way to get a much higher quality out.
24:09
Daniel Scharff
All right, Chris, these are amazing tips about AI in general. I feel so much more equipped to sound like I know what I'm talking about AI. What are some specific use cases now for all the CPG founders out there?
24:22
Chris Szigeti
Yeah, I'll just give a couple that I use like either weekly or just have used before as I start to build out my brand that I think are really helpful. The first one is, and we didn't really touch on this earlier, but there's kind of another type of model in these foundational models that's called like a Deep Research. So Google has this in their Gemini model. ChatGPT has this. If you pay for the $20 a month subscription, which I would say just pay for the subscription, it's totally worth it. These Deep Research reports, I think are game changers for brands, right? Like, I used to do a ton of desktop research. You know, I would go and scrape McKinsey reports and you know, industry reports.
24:58
Chris Szigeti
And now what I'll do when I want to learn about something, like a really good example of this was like the honey industry. I wanted to know stats about what's the growth of it, you know, how fast is it growing, like who are the big players? And so I ran a deep research report in ChatGPT. I asked it some of these questions. And going back to our prompting tips, the way these deep research models run is they actually always will ask you clarifying questions to make sure that they can get the best output. So it'll ask me a bunch of clarifying questions about honey, space, my brand, et cetera. Hey, do you care just about the US market? Do you care about the global market? Things like that. And then it'll actually go off for like sometimes 30 minutes to an hour.
25:36
Chris Szigeti
And then it will compile a research report. And so the one that it did for me was like 25 pages or something very akin to like what you would see in again of the consulting or industry report. It'll cite its sources, which is great. And there's kind of two really useful ways you can use that. One is, hey, you just got a lot of background and you can deep dive into this. The second is it'll actually give you like a table of all the sources that cited in this. And so instead of me having to go out and do a bunch of googling, find these industry reports, I can just read the report, read its summary, and then I can deep dive into the sources.
26:09
Chris Szigeti
So this has been really useful as I built my brand for market research, you know, competitive research, even things like I ran a deep research the report the other day like just the 3 PL industry as a whole, right? Like, what's a good way as a CPG brand to utilize this industry so you can run these deep research reports? It's super easy. It's super simple. Saves you a lot of time and headache. Again, you got to fact check some things. And that's why I like these deep research reports, because it cites all its sources. But that'd be kind of tip number one is it's easy, it's out of the box. You just get a paid subscription and you can go use these for research.
26:43
Daniel Scharff
All right? And before we go to tip number two, let's just acknowledge sometimes like the same way if you just googled an article like, how do I grow my Velocity, It'd be like 1, 2 and 3. It actually might not be true. Like there might. Then point number two might not apply to your kind of a product. So you know, yeah, look at the sources and think about it. I personally use it to get ideas, but also you have to rely on your judgment to know what's good and bad. The same way if you were talking to somebody, they're going to tell you based off their experience and their journey, most of the time not based off as much your journey and your product and maybe not even knowing what could best in your situation. Right?
27:22
Chris Szigeti
Totally. And hey, it's trained on the Internet. Not everything on the Internet's true, right? And that's where if you can provide additional context, right? If you have relevant research reports or relevant information, like always give it that because it's going to use that information. It'll get a better output and response. And another tip too that I did not mention earlier, but you can actually specify like how you want these models to respond to you. I think you do it in the settings in ChatGPT, I'm not sure in other models, but it's usually just an input box. And so what I have in mind is like always respond with data in tables, right? Cite your sources, lean towards like academic sources and relevant research sources compared to like blog articles, for instance.
28:04
Chris Szigeti
So again, you can specify to try to steer it in the direction of using those higher quality sources. But that is one of the benefits of the research report or the deep research models is you can dig into those sources and you can find a lot of good stuff you wouldn't find otherwise.
28:18
Daniel Scharff
Okay, perfect. Tip number two.
28:21
Chris Szigeti
Tip number two. So this is a fun one actually that friend of mine who founded a tech business told me, which was you can actually use like these models to be like a virtual advisory board, right? And this is one thing it's really good at. So I'm building a small company. I do have some advisors on board, but there's a lot of people that I can't talk to every day. So what I do is like once a week I go on like what I call like an advisory walk. And I just tell ChatGPT like, hey, you're my advisory board, right? There's five of you. Like one of you is a go to market expert, one of you is a branding expert, you know, one is an operational person, you know, whatever else you might care about that week, right?
28:59
Chris Szigeti
So you can give them kind of these Personas almost, right? Or these roles and then you can go and what I do is I actually use, like, the talk feature on it where it will talk to you. It's a little creepy, but it's. It's really good. And I can multitask, so I'll go on a walk and I will just work through, like, whatever problems I have that week, right? Like, I might have a branding problem or a trademark problem. I mean, I can just talk through it and get advice from it as if it was like a virtual advisory board. It is a little dystopian. It's a little weird and creepy. And again, you know, not everything it gives is going to be perfect, but I can actually talk through these challenges in real time.
29:33
Chris Szigeti
I can ask it, like, hey, you know, based off what you know about my company, like, what should we focus on next week in, like, the ops department? Right. That might be a bit general, but you kind of get the. Get the gist.
29:43
Daniel Scharff
Give me an example. Because, I mean, that sounds creepy, like you said, but also I'm like, wait, is it good, though? Because, I mean, maybe because I'm doing it wrong, but when I'm putting some prompts into chat GPT, like, it takes me a couple iterations and a bunch of like, crap before I get to the thing where I'm like, oh, that's good. So, like, I feel like because I can read so fast, everyone, you know, we read much faster than we can actually process words that it's okay because I can just scan stuff and I know where it is. I can do it really fast. I'm like, that would drive me crazy to be having a live conversation that way, where I'm just like, no, no, do it this way, do it that way.
30:16
Daniel Scharff
But it sounds like, I mean, if you're still going on these walks, obviously it's productive for you. So is it better than that? And because, you know, you've learned how to use it and. Yeah. Do you have a specific example of talking to them about something?
30:28
Chris Szigeti
Yeah, I mean, I think it's good for use cases like this where you're like, hashing things out or talking through things. Right. It's not as good, in my opinion, for being like, hey, I have this. I have a question on specific thing, or I want you to research a specific thing. Like, it's good when you're like, multitasking. You're like, hey, tell me about the supply chain of. In the honey industry. Like, where do things source from? You know, how does that work? And it can just kind of give you some research. You can do some Back and forth. So it's good with things like that. But for the, like, advisory kind of board model, like, the one thing I do is one, I define the roles at a time. So I say, like, you know, these are kind of the makeup of this.
31:04
Chris Szigeti
And then the second thing I do, and this is another kind of tip and trick is I'm a big fan of like, pitch decks, right. Even if you're not raising money, I think it's really good because it forces founders to consolidate their thoughts, be able to communicate their business model, et cetera. So I always have a pitch deck for both the companies I build and the companies I work with. And pitch is a bit loose. It's a big strategy deck, but it can also be a pitch deck. And what I do is I always keep this up to date. So when I have these conversations, you know, I just give it the latest version and I say, hey, here's a bunch of info about my company. Here's the roles you are and here's the problems I have this week, right?
31:40
Chris Szigeti
So like last week, for instance, I was having some issues with like a packaging supplier. You have a unique packaging. Can't get it in the U.S. It's, it's very complex to get. And so were talking through some options of like, how might I look to source this type of packaging elsewhere, right? Like, it was like, hey, have you considered looking at, you know, these countries are working with a packaging supplier that's not getting it directly from the factory or, you know, give me some examples. Again, a little generalized here, but that is like kind of the sample of bringing a real problem and trying to talk through it. I'm not just going to take what it says as gospel, right? Like, you know, I'm going to go do research after and have further conversations about it.
32:21
Chris Szigeti
But that's kind of how I use it in that advisory board model. And again, I just do it once a week. I just talk through the challenges I've been having as a company that week, get some feedback, some thoughts, and it helps you think about perspectives you might not have considered otherwise.
32:33
Daniel Scharff
So it might be good at asking you questions, right, and helping you think through stuff. But then hopefully, like a good advisor also is going to have experience where they can also say, well, here's how some other companies have done it. Does that apply to you? Or, you know, help you think of stuff like that? And then, okay, so I like that part also about using it to give yourself customized research reports, right? Like, sometimes if I'M just even traveling to some historical site or country. I'm like, oh, I should. Let me find a podcast real quick on that. And I can just get up to speed a little bit. But it's like that. But you can do it customized and it'll talk to you. Are the voices cool or are they weird?
33:11
Chris Szigeti
You can customize them. They're a little weird, to be honest. It feels a little like early days of Siri almost. But again, it's a good tip. Like, I don't need to be sitting down and just typing back and forth. Like, I can have more of a conversation with it. Got to get over the hurdle. It is a little weird. It's a little dystopian. But it's a good way to hash things out and talk through them within real time.
33:34
Daniel Scharff
Hello, Chris. Do you think you could do an impression of it for me? Is this good?
33:41
Chris Szigeti
That's pretty good. To be honest. I think you're missing the accent. I think Myrna has a British accent. I won't make it up. Hello, Chris. That is a bit closer. That's a bit closer, for sure.
33:51
Daniel Scharff
Can I give it. Yeah, Can I give it a cockney accent? Can I. My mom's Scottish, so maybe I could give it a nice Scottish accent. That would actually be very soothing for me. All right, let's. Come on. Let's not be so silly here, Chris. This is a serious topic.
34:05
Chris Szigeti
All right?
34:06
Daniel Scharff
I'm talking to myself here. Let's keep going here. So. All right. It's like, I just. It seems like it'd be uncomfortable to listen to that, but, like, you think general people are open to that kind of a conversation to have, or is it just because you're such a tech head that you, like, really want to adopt it and you're already okay with it?
34:25
Chris Szigeti
I think it's both, to be honest. Like, I definitely am on the early adopter curve with a lot of this tech, right? Like, I use it at other companies. I use it when I'm consulting and advising. I'm building products in the space, so I definitely try a lot of the new stuff. This is why, going back to what we said at the start, I'm really big on, don't get analysis paralysis. Just start using this stuff, right? You might use this advisory kind of, or talk to text feature, for instance, and be like, that's not for me. That's great. There's plenty of other ways you can get value here. So I do think I'm the front of this journey. I think it will Take a lot of other people time.
35:03
Chris Szigeti
However, what I'll say is I work with a lot of folks who are not super tech savvy, and they start using these, and they're like, wow, this is really beneficial. Right. So it all comes back to value. I think if you get enough value from it, you'll get over kind of the roadblock of, hey, this thing is kind of creepy. It's, you know, it's talking to me. It sounds a little robotic. And again, it'll get better and better. The way I think about it is, like, if you start using these tools now, like, one, you're gonna get some value from them, and two, they're getting better so fast that, like, you know, the longer you wait, I think the harder it's gonna be to hop on the train.
35:37
Chris Szigeti
Even though they're getting easier to use, like, up that muscle memory of how to use them. You know how to use them effectively. You know how to tailor them to your specific use cases. And so it just takes practice.
35:48
Daniel Scharff
Okay. And then just to be very tactical about this, how do you use the voice mode? I don't. I have chat GPT. I don't know how to do that.
35:55
Chris Szigeti
Yeah, there's a little button, and it. Right when you're typing in on your phone, for instance, next to the microphone, there's like a little audio signal button. You just click that. It's a white button, and it pops up this creepy blue orb, and you can talk back and forth to it. You might have to have the $20 subscription. I forgot what they call that. The team or plus or whatever it is. Don't know if they offer that on the free plan, but again, as I mentioned, like, just pick one of these tools, pay for it. Like, they all have pretty similar features, and it's worth the $20.
36:25
Daniel Scharff
Okay, that is cool. And it's right there, hiding in plain sight. It's a little black and white thing. It kind of looks like the sideways Spotify logo. And it says, use voice mode.
36:36
Chris Szigeti
Exactly. Yep, right there.
36:38
Daniel Scharff
All right. You're blowing my mind right now, Chris. I love this. Okay, tip number three.
36:44
Chris Szigeti
Tip number three. So, and this is pretty similar to this advisory board one, but also just when you're prompting or asking, you know, a specific question of these models, like, you can tell it, hey, you are, you know, a sales guru, right? Or you are a revenue generation expert. Like, giving it that role or that Persona, you know, it's essentially, you're just giving it extra context, but you're telling it very specifically, like, Think like this type of person and then I'm going to ask you a questionnaire, right? And the way I like to approach this is for specific questions. You know, if I have a revenue question or a go to market question, I'm gonna say, hey, you're a go to market expert in the CPG space, right? Specifically for companies, you know, under a mil in revenue, because that's where I am.
37:26
Chris Szigeti
So, hey, think through that lens and then I'm going to ask you a question. So just giving it that kind of Persona I think is really helpful as well. Something not a lot of people do. But if you do use some of these more advanced tools and you actually look behind the scenes, that's kind of what they do behind the scenes, which is cool. So that's just a tip for the everyday person, is like, give it some specific role if you have a question. You know, if you were going to go ask your CMO something and you want to ask a marketing question, say, hey, you're a CMO for a CPG brand and the response you get will be a lot more tailored to them.
37:54
Daniel Scharff
Okay, so on just on tactically on that topic, I have chat GPT. Would I. I've heard some people talking about it as if they're like different bots that you would create. Like, okay, let me talk to my marketing wizard. Let me talk to my. Are these like separate instances? Because as far as like in my chat jbt, it just seems like there's just one prompt and I'm just having one long conversation with it. But like, do you have separate threads that you use? Help me.
38:18
Chris Szigeti
So I, I open a new thread every time. I have like a different kind of train of thought, right? I don't do it all on the same thread. The ChatGPT, all these tools, they have like what's called memory context. So if you talked about something six months ago with it can call back to what you talked about six months ago. So you don't need to be super particular about organizing and structuring it. Just get the content in there and it's smart enough to extract it. But in terms of these different bots or agents, you'll hear the term agents thrown around in the AI space. There's all sorts of products where you can create like a marketing bot, right? Or, you know, a newsletter bot or whatever it is.
38:54
Chris Szigeti
But when you're interacting with just these ChatGPT, these foundational models, just do it right in there and just essentially treat each new conversation as like a different thread and Just say, hey, you are a blank. You know, this is what I want to do. You know, help me out.
39:07
Daniel Scharff
Okay. So, like, there are projects I guess you can create, but that would just be for your benefit to keep your, like, so you remember where all the questions you asked were in case you want to scroll back through. ChatGPT doesn't care. They know everything that you ever talk about, so it doesn't have to be organized.
39:20
Chris Szigeti
Yeah, it didn't used to. It used to only have a shorter, what they call a good context window, so we'd only look back a little bit. But now they've expanded that. I don't know if there's a limit on it, to be honest, but I don't think most people are going to hit it. Just have the conversation in there. And this is also why I think it's valuable to just pick one of these tools instead of using three or four. Like, if you pick one, you're putting all your specific context memory files, et cetera, within it. So you don't need to be, oh, I used, you know, Gemini yesterday, but it doesn't have my, you know, pitch deck I put in last week. So it's easier to just centralize, in my opinion. Use one, master it and learn it.
39:56
Daniel Scharff
Okay, so, Chris, going back to that meme thing, like, where you can create images, with ChatGPT, it can kind of do whatever it would. I use ChatGPT and tell it to create an image. For some reason, it always looks like a 70s circus poster or something, no matter what I tell it to do. So I don't think I'm doing it right. But, yeah, talk me through the image generation stuff. What's possible? What are some tips?
40:16
Chris Szigeti
Yeah, absolutely. So, I mean, this is where we've seen a lot of the advancements in the last couple months, to be honest. Right. ChatGPT released their new 4O image model. Again, terrible name, but this is the one that broke the Internet with the memes and everybody, you know, making their pets into humans and all that fun stuff. So these image models is where you're seeing a lot of, like, they used to be pretty gimmicky, to be honest. Right. Or they wouldn't work great. The advancement here is just. It's going through the. So, you know, a couple months ago, I would have said, like, hey, these are gimmicky. You can't use them a ton other than maybe brainstorming or kind of conceptualizing stuff. But now you're starting to see people use these image.
40:57
Chris Szigeti
And to some Extent the video models, but they're not great yet still. But the image models especially like in their business, in their day to day. So if you're using again, I'll just Talk about the ChatGPT one but their 4.0image model. The way you use this in the tool is you just say like, make me an image of whatever. It knows it'll use that latest model and you can, you know, prompt it, you can go crazy, you can do all the fun stuff. But a couple really useful things you can do with this in the CPG space. One is conceptualizing, brainstorming, et cetera, right? So you can say, hey, you know, I'm thinking of switching my packaging to, from a, to a glass jar. Like here is my branding logo, colors PDF. Like mock that up.
41:38
Chris Szigeti
It's not going to be ready to go take to a packaging supplier and say hey, give me this. But it's really good for conceptualizing, like is this a direction I want to go? Like does this look good? Okay, what I want to change, you know, the color of the label or something like that. So it's really good with conceptualizing, brainstorming, et cetera. You may have to do a little back and forth. Another, you know, thing. I see this a lot in the slack. If somebody says, hey, I uploaded a picture of my product, I made a social image using it. So I uploaded a picture, I said, hey, you know, here's a jar of honey. Put a bunch of bees in the background and make this, you know, for Instagram.
42:12
Chris Szigeti
And it will go off, it'll do its thing and then people will say like, hey, it changed the label a little bit, right? And the reason for this is because it's not taking exactly that picture. You input. What it's actually doing is it's taking that picture, it's regenerating it based off what it knows and then it's adding in the extra context. And so this is why it's still a bit challenging to use these models for like really good like you know, advertisements or imagery that you want to put because it might change something a little bit and it's really hard to prompt them to get an exact one to one. It's getting a lot better.
42:44
Chris Szigeti
I think very soon you're going to be able to just put in, you know, mock ups of your product or whatever and then say, hey, make an advert for this, you know, in this size, for this for socials or for Google Ads or whatever. And it will spit back something that you can actually just go and use. But it's getting a lot close. So conceptualizing, you know, getting a rough draft. Like the way I use this is if I want to create like a social ad, I'll mock it up using ChatGPT. I'll describe what I want, I'll kick that over to a designer who then will go put like the last 20% on it. Right. Or they'll just say, okay, I get your vision, I get your direction. Like, let me make a real advert that you can go use and feel confident in.
43:24
Chris Szigeti
So that's kind of one. One way you can use these image models. You know, there's a whole bunch of different ways I've seen people make really cool GIFs with them for, you know, socials. I've seen people, you know, use them to create images for blog articles. Like, those are the things that are maybe a bit lower risk. That's pretty good at doing. But again, my kind of prediction is within a couple months, you're going to be able to use these image models right out the box to do a whole lot of really interesting things in your business.
43:49
Daniel Scharff
Well, I look forward to that. Although, I mean, our startup CVG Designer, is the best in the business. Everything that he does is incredible. But it will be nice when in the middle of the day if I want to just like real quick use his kind of guidelines for a brand and dump out a new image for an event that we're doing. That will be fantastic. So I. Yes, that will be very cool. And you know, it's. It's kind of funny when you say that. I just almost like, for someone like you, sounds like you know what you want a lot of the time. Yeah. If you can have it give you that. It helps you communicate what you want to a social media expert or designer, whoever. Like, this is what I'm going for.
44:27
Daniel Scharff
Because, you know, we all know with words, they mean so many different things to different people. So if I say dog, like, tell me what's in your head right now, Chris.
44:38
Chris Szigeti
Exactly right. I picture like a cartoon dog was the first thing that pops into my head. And that's probably not what you were looking for.
44:44
Daniel Scharff
Yeah, when I said dog, I actually was picturing a real live golden retriever and I have a dog and I wasn't even picturing my dog and other people. It's just like. And. And first. But somebody else who's afraid of dogs might picture a menacing dog. And it's so different based on the.
44:57
Chris Szigeti
Life that you've lived.
44:58
Daniel Scharff
Right. And just having an actual Image that's a little bit further can save you probably a lot of time when you're, you know, they have that whole, like, house tree diagram test for kids, and they, like, use it to, you know, see how. How, like, confident are you in yourself and, like, whatever family issues you might be having, like, it just. You can learn a lot through that stuff. But also it just really can help to try to get more specific when you have an idea. But also a lot of people don't have that idea. And if you actually try to give a more specific idea, you could also actually limit the creativity of the person that you're working with.
45:31
Daniel Scharff
Because if I give our designer, like, hey, this is what I want, then I might actually not get the chance to learn about some incredible new idea that he would have. That's better for sure than anything that I would think of.
45:42
Chris Szigeti
Yeah, absolutely. I mean, it's a communication tool, right? And that's one thing you can really use these image models for, is communication. But I agree, like, I don't think designers are going away. Right. I think they're going to be the ones really adapting and utilizing kind of AI and the tools to get better output, you know, faster, but just be able to do things you couldn't do before or too time intensive or, like me as a small CPG brand, like, I might not have the money to pay a designer to go do a humongous project, you know, social project, but if they can do it in a fraction of that time, right, like, great. Or if they can just do all sorts of stuff. So I think the.
46:18
Chris Szigeti
The way to look at it is not like it's going to replace a lot of these roles and functions, but it's going to kind of change them, augment them, and just improve the quality of.
46:26
Daniel Scharff
Yeah, I mean, think of basically any post you ever do on Instagram. In the olden days, that would have required a team of video editors and graphic designers. And now we just do it with a couple swipes and it looks good and we add music in, and that would have been a whole mixing exercise. So, yes, that makes a lot of sense to me. Okay, so another question that I have for you then, Chris, maybe to kind of like, sum up all of these tips. You're telling us all about the things that we could do with AI, but you know, all these things and you are actually growing your brand. So can you tell me specifically how you have been using the AI tools to grow and how you anticipate to continue to use them just to make your Brand better, stronger, more efficient.
47:05
Chris Szigeti
Yeah, totally. So I'm a solo founder, right? It's just me on this team. And because of that, I use, you know, contractors for a lot of work. And you know, five, 10 years ago, like I might need to build a really robust team around me and do a lot of the things I couldn't do. With the advent of all these AI tools, I now can do so many things in my brand and just accelerate like the development cycle. The time it takes to build a brand and go zero to launch, you still have supply chain issues, you got to make the physical thing, but all the behind the scenes stuff, it's just accelerating. And so I've used AI in pretty much every facet of building this brand from day one, from everything from my, hey, I have an idea for this product.
47:49
Chris Szigeti
Let's talk through it. Does anybody do this out here?
47:52
Daniel Scharff
Right?
47:52
Chris Szigeti
Going back to the competitive research, like, is there a product like this on the market? Hey, I don't really know the honey industry. I'm not from it. Like, tell me about it even to like, hey, I don't know cpg. Like, what are the steps that it takes to go from a product to go from an idea to that first launch? Right? Like, how much capital should I have up front? Like, things like that? So the very early on, just like going from idea to almost like business plan was very heavily influenced from deep research reports, prompting back and forth again, not just taking what it says as gospel, but even knowing the right questions to ask. Right. Like, I had heard of 3pls, I'd never worked with one before.
48:29
Chris Szigeti
So you know when it's like, hey, if you want to ship product to your customer, you can ship it yourself, you can use it. 3 PL here's different options. So it really gave me the information to just know what questions to ask when building a business. And if there was something that I was confused on, I would just deep dive with it. So that's, that's kind of one way. Another way that I've done it was using those image models to really take what was in my head and like put it on paper. So I had an idea for a product. I knew the packaging had to be really unique and stand out. And so I just went through iterations of like, hey, what are some really unique packaging types that we could package a honey based product in? Right.
49:06
Chris Szigeti
Like, what are most honey products packaged in? You know why, Right? Like, I think a lot of people don't ask the why, like, why does it always come In a glass jar. Right. You ever thought about that? Or like why does it have that bare squeeze bottle? Right. The questions you just don't even think to ask. And so asking those and then using it's okay, well let's brainstorm some unique packaging options that will really pop on the shelf. But you can still ship D2C. So I've done a lot of that type with it and I've even gone as far as again, a lot of people won't do this but like creating the entire storefront which isn't launched yet. I created that all. I'm not a developer, I can't code. I created it all, you know, using a tool called V0 and hooking into Shopify.
49:46
Chris Szigeti
So because I wanted to have a really fun unique one that I couldn't just do out of the box with the Shopify theme and going back to small team, didn't want to hire somebody pre launch to go build this for me. So I went and built it myself. I don't expect most people to do that, but I think that's the direction that a lot of these tools are trending where it will, you know, somebody who's a solo founder or non technical, you know, they can do things that they might have to pay, you know, thousands of dollars prior. Some they'll just be able to do themselves, some they'll be able to get, you know, 50, 60% of the way there and then tag someone in to Finish it.
50:16
Daniel Scharff
Okay. V0.dev I think is what you used, which you just know about that tool or you had googled to figure out how can I do a Shopify store using AI?
50:26
Chris Szigeti
Yeah. So there's a, I mean I know about it is the real answer there is kind of this category of products popping up which are like web development tools. Right. So if I used to, a year ago, if I wanted to build a website, I would either use a Squarespace or a Shopify. You know, one of these editors and I would make something, some you can customize a ton, some not so much or I'd pay somebody to go do it. You know, now you're having these tools pop up that allow non technical people and I'll put non technical and air quotes for the people that can't see because you deal do still have to have a little bit of an understanding of how to use this stuff. But, but you can build essentially websites from scratch.
51:03
Chris Szigeti
You know, you can launch them, you can put them out into the ether so anybody can go access them and I think we're a couple months away with tools like V0replit's another you might hear of, or lovable. These are three different, just like web development kind of code creation tools where I can prompt it to do something like, hey, make me a website for an emerging honey brand. It'll just take what it knows and spin one up and then I can customize it and do all that stuff. But I think we're a couple months away from really anybody being able to use these to create like websites for instance. So it's just this effort tool. It's, you know, it sits on top of those foundational models like I talked about.
51:41
Chris Szigeti
But you can do some pretty cool stuff if you're willing to play around or have any essence of, you know, technical abilities.
51:47
Daniel Scharff
Yeah, I think for me I'd be like so afraid, like, oh, I'd do that. But then it would some one thing would go wrong and I would have no idea how to fix that. You know, like the same way I'd be like, oh, I can fix the toilet. And then one thing would go wrong in my house we flooded and I'd be on the phone with a plumber.
52:01
Chris Szigeti
Well, that's what's really interesting about these tools. Not to nerd out too much, but when they first launched, that's exactly what would happen, right. It would generate the code. If I had an error, I'd have to go manually troubleshoot and fix it. Now when you have a bug or an error just pops up a red box that says fix with v0, you click that button, it auto tries to fix it. And I bring this up because you can see the direction these products are going. Where they start out. You have to be pseudo technical or at least have some domain knowledge to use them. But very quickly they're going to get to a point where like just normal, you know, CPG founders who might have zero tech experience can use them and get some sort of value out of it.
52:38
Chris Szigeti
So that's really exciting, I think.
52:40
Daniel Scharff
Okay. And then anything else you foresee using it for, as you grow your brand, you're on your way to millions in revenue. Maybe you're still a solo founder, who knows? But any big plans that you have for using AI in your journey, I.
52:55
Chris Szigeti
Think I'll use it as long as I can and for everything I can. To be honest, my plan as a solar founder who's also bootstrapping is be as lean as possible and wait to outsource until I have to. So if I can do the first six months of social posts, just using AI to help me generate social post ideas and write the content and maybe generate some of the images or maybe I outsource that. But if I can use it to do a lot of that until it becomes too overwhelming to where I need to outsource or hire somebody to do it, like I will. And that's just what I think is a really interesting time to be at this intersection of both CPG and AI because it's still very capital intensive business, right?
53:37
Chris Szigeti
It takes money to buy packaging, it takes money to buy and manufacture product, to ship it. Like you've got to have the capital to do it. But I think a lot of the overhead items that solo founders either would have to raise pre launch or you know, just pay tons out their pocket to do, I think a lot of that overhead is going to go away into these AI tools. You know, there's a bit of a learning curve. You know, I'm not saying it's all going to go away and you can just launch a company, but I think a lot of that capex that you needed up front, you're going to be able to stretch a lot further. So I plan to use it for everything.
54:08
Daniel Scharff
I gotcha. I think probably a marketer would have a different perspective on this. But I can say like, for brands that I've done in the past, oh, we wasted money on our Instagram where it's because, you know, like, I think if you haven't done it so much, you just look at it Instagram and you're like, well, I need to grow my Instagram followers and then I'll have a big brand and everybody will buy me and all the stores will want to have me. And that's just what you do, you know. Then was actually looking around at competing brands. We're like, whoa, this brand has been around for 10 years and they only have like 20,000 followers even. But they're a real legit brand that is nationally distributed with high velocities.
54:44
Daniel Scharff
Like oh, maybe only after we had dumped a bunch of money into not getting followers, where we like, maybe we shouldn't actually spend all that stuff. So but some people do discovery and search for products on Instagram at the same time, even if they're not going to follow you. Like, yeah, buyer might look or consumer might be like, oh yeah, let me check their Instagram and see what that looks like. And yeah, we had spent money not just promoting posts, but actually creating the posts, like having a photographer go out and create Imagery and do all this stuff and, you know, paying for, if you need a model or something for the photo or just, like some assets for it or whatever. Those expenses are kind of crazy when you think about the following that most brands will have in the early days.
55:24
Daniel Scharff
Now, I wouldn't say that creating AI images is going to go a long way toward actually building community around your brand. And, like, they're obviously the authenticity of the message and all that stuff. It's, you know, it may not get the job done on that, but I don't know, it could actually luck out and create an incredible ad that then you want to start promoting. So I sort of like it for the, you know, the cost benefit, I think, especially on a lot of brands that are going to be more retail focused that just you kind of need it like you need a website, but you don't necessarily need to spend so much money on it because you could spend that money on slotting fees.
55:59
Chris Szigeti
Yeah, I totally agree. And I think the trend we'll probably end up seeing is not that AI replaces all this stuff, but I think it's going to let people pick and choose what they want to focus on, what they can do themselves and what they want to bring experts in, Right. And going back to your example, you know, maybe an early brand's like, hey, I can do a lot of the Instagram stuff myself now, right? Like, maybe they have a bit of photo background and with AI, they can create the content and they say, I'm going to take that, you know, five grand I was going to spend on that. I'm going to push that into another bucket, right? And they might see more ROI there.
56:30
Chris Szigeti
Or the other thing that you see in tech a lot is it actually lets you run more experiments, and then once you find the one you want to double down on, that's when you go deep, right? So you might push out four or five different types of, you know, content or blog post or whatever it might be using kind of these lean AI tools. And then when you find one that gets a little traction, that's when you're like, all right, you know, I'm going to put a little gas on this fire and bring somebody into to really do it. Right. At least that's, you know, that's a lot of what I'm using and seeing. And then I would also say, I totally agree with authenticity. I think this is a big challenge.
57:01
Chris Szigeti
I would not encourage people to go out and just create a bunch of AI content and push it out into the world, because consumers are smart and consumers will see right through that. Somebody told me once, and now I can't unsee it, that you can always tell AI text because it uses a lot of dashes, because the Internet uses a lot of dashes. And once you know that, you can really pick out AI content because that's just not the way humans write. Right. Like I don't put in tons of dashes everywhere. So, you know, be authentic. You know, don't just use it as a spam tool. Right. But use it as like an aid in your work is my advice.
57:35
Daniel Scharff
Yeah. And I mean, it is amazing how much it can do. I originally, I think in the early models I would see a paragraph, I'm like, I know AI wrote that. That is some bs. Like it just uses crazy language that it just doesn't sound like a normal person wrote. It's not interesting writing. But now, gosh, it's crazy the stuff that it can provide. I've given this example before, but I had to create a job description for me and it was so much more me than what I had written. It was, it just had written better in my voice than I could. It was super impressive. And then that just saved me the hour that I would've spent actually writing that so that I didn't change a thing. It was perfect. It was so good.
58:11
Chris Szigeti
Honestly, I think content and tone is really important, right. So I always encourage people, if you're gonna put something out externally that AI generates, like write the first couple yourselves, right? Like blogs, Write the first 10 yourselves. Put it in a chatgpt so it knows your style, tone, and then don't just like take what it spits out and then put it out there, like go through it, right? Be the human in the loop. Like check it, make sure it's good. But the more info you can provide on how you write, you know who you are, you're going to get a much better output.
58:38
Daniel Scharff
I think that is very important. It reminded me somebody told me this story, they're like, oh, I actually used AI in my job performance review to do with my, you know, self submitted review. And it pretty much ruined my job because they said that actually like the marketing person that they reported to was all about AI in the early days and all this cool stuff. And so, you know, everyone hates these self evaluations that you do every year. And so when they got theirs, they just used chatgpt or something to do it and then it. But it used such over the top kind of language. Like, Chris is a landmark contributor to the end drove the foundational strategy behind how we use. And it was just so over the top that it really pissed off this person's boss.
59:26
Daniel Scharff
And, and he said their relationship was never the same even after he was like, oh, I'm like, I'm sorry, I. This thing is stupid and I just use chat GPT to do it because I don't care. And like, kind of thinking maybe they would respect that and they did not and they were pissed forever. And I think this person had to leave the job basically. So human loop feedback important.
59:46
Chris Szigeti
Yep. Yeah. Hey, I mean, at the end of the day, business and startups, it's people problems, right? Like everything is, you know, people interaction. So you gotta be careful because as I said, people are smart, they can see through things. Authenticity is super important. Like, and again, also, don't just trust everything that comes out of these AI bots, right? Especially like with math and calculations and numbers. Like the models are good, they're not nearly as great as they are at generating text content. So if you're going to make a big decision using an AI, you know, calculation or something like, make sure you fact check it.
01:00:20
Daniel Scharff
All right? You hear that, AI bots? We're watching you. Don't watch me, please. Okay, so Chris, just take us out here. One last question there. I know you know about a bunch of these different tools like Veo dev and stuff. Tell me about some others just so we all know what we don't know that's out there.
01:00:37
Chris Szigeti
Yeah, just a couple other kind of ones that may or may not be useful to people. But again, I mentioned Perplexity. That's just kind of like Google AI. It's like a search engine, right? It's really good. It's great for, you know, finding products. I think they even introduced a shopping feature recently that I haven't played around with a ton, but I've heard good things about. So Perplexity is one for search. Gamma app or just Gamma. This is a personal favorite of mine. I love slide decks because I use them for everything. I hate creating them. Gamma is what I've always wanted AI to do, which I can take an outline of a deck, I can throw it in there and it will create a slide deck out of it. Again, it's like an 80, 20.
01:01:14
Chris Szigeti
I go in and I massage the last 20% of that deck. But it's really good if I want to create a custom deck to communicate something, for instance. So Gamma app I love. Canva is kind of doing some similar stuff as well with their AI. Features I already mentioned v0 which is like UI web development. Similar ones to v0 would be Replit love. They all kind of do the same thing. But I think Dzero is the best one. We haven't talked a lot about but like meeting notes and transcriptions. So I used to use Zoom for everything. I now use Google Hangouts or Google Meet, whatever they call it. Because one, the audio quality is getting pretty on par with Zoom. But I just think the auto Gemini note taker is fantastic. Like I don't take notes in meetings anymore.
01:02:01
Chris Szigeti
I let that take it. I focus on just the conversation and the meeting and then I'll take the meeting summaries after. They're pretty good. It has the raw transcript. A lot of times I'll even take those and put those meeting transcripts into ChatGPT and ask it more questions about it. So Gemini for meeting notes. But there's other tools like Otter has been a popular one. There's a million AI note takers. Pick the one that works for you is my advice. And then lastly for visuals like Mid Journey is another one. I haven't played around tons with it, but I've heard really good things from CPG brands. It's built on top of the OpenAI model for ImageGen, but it's a bit better. It has some improvements as well. So you know there's a million tools out there.
01:02:41
Chris Szigeti
My advice to people, again, don't get analysis paralysis. Don't go trying every tool. Like start with a problem. When you have a problem, go see if there's a tool for it. If not. Right? There might not be one yet, but that's the approach I would take because you really can get lost in the rabbit hole of these things.
01:02:59
Daniel Scharff
Do you think Gamma that does the slide decks can work with your custom branding deck and create things that will feel like your brand. Or is it doing more basic template stuff?
01:03:10
Chris Szigeti
It can do a little bit of both. So they have a like a theme. So I have my brand themes put in there, right. Colors, fonts, logos, all that stuff. I think again, I think Canva does the same thing and it will generate it. The way that these like slide deck tools work, it's like they kind of have a standard like template framework, right? Like text boxes and you know, visual layouts. And so what it does is you give it an outline, it will structure the deck based off the content you gave, plug it into one of its kind of template things and then if you gave it, hey, use this brand colors, it will do that. So Again, I don't think those are very close yet to like, I can just put in an outline and have a hundred percent ready deck.
01:03:50
Chris Szigeti
You still got to do some massaging, but I think it takes out like a ton of the work that you do when creating presentations or visuals there. So you can go and say, I need a presentation to communicate. X, Y, Z. Create me an outline. I can put that in gamma if I want and have it do a lot. So it's just streamlining a lot of the tasks. And I would say this is like, this is the common theme with AI tools is I don't think they're going to replace jobs, you know, maybe some, but I think what they're really going to do is replace that 50% of your work or job that was kind of mundane and menial. Right. Like just writing down an outline for a slide deck and then going in and tweaking boxes.
01:04:27
Chris Szigeti
Things that like, I don't need to be spending all my time doing that.
01:04:30
Daniel Scharff
Right.
01:04:30
Chris Szigeti
Like, I'd rather do the higher order stuff. And so that's where you're seeing a lot of the convergence around these tools.
01:04:36
Daniel Scharff
All right, Chris, this has been quite a journey. Thank you so much. I feel much more knowledgeable and better equipped to deal with AI conversations, especially at parties that happen. I'll sound like a pro now, as will everybody who has listened to this. So. So Chris, what is a good way for people to follow along with you and support your brand as well?
01:04:56
Chris Szigeti
Yeah, you can follow. You can just go to frothinhoney.com you can sign up for our newsletter. I will keep posted updates, all that good stuff. If you guys have questions on AI or any of that stuff, I'm in the Slack group. You can just message me, you know, hrisaghetti. I'm always happy to have these conversations. I love talking about this stuff. It is something I nerd out on. I know a lot of other people do. But if you're just, you know, lost in trying to get started, have a couple questions like just hit me up in the Slack channel. I'm always happy to have conversations around it.
01:05:26
Daniel Scharff
All right, Chris, thank you so much. Hope everybody enjoyed. See y' all later. Thanks, Daniel. All right, everybody, thank you so much for listening to our podcast. If you loved it, I would so appreciate it if you could leave us a review. You could do it right now. If you're an Apple podcast, you can scroll to the bottom of our Startup CPG podcast page and click on Write a Review. Leave your company name in there. I will try to read it out. If you're in Spotify, you can click on about and then the Star Rating icon. If you are a service provider that would like to appear on the Startup CPG podcast, you can email us@partnershipstartupcpg.com lastly, if you found yourself grooving along to the music it is My Band. You can visit our website and listen to more.
01:06:12
Daniel Scharff
It is superfantastics.com thank you everybody. See you next time.
Creators and Guests
