From ChatGPT to Local AI: Practical Tips for Marketers with Christopher Penn

Christopher Penn [00:00:00]:
If you’re a marketer and you’re like panicking about what to do, do a good job with SEO and also make content people actually give a about.

A. Lee Judge [00:00:09]:
Welcome again to the business of marketing. I’m a Lee judge. You know, in most circles that I find myself in, I’m typically one of the more knowledgeable about using AI for business purposes and even creative purposes. But more than a few times I can be quoted as saying I can’t teach AI when I know people like Chris. I truly realize how much I’m a newbie to AI when he’s in the room. And for that reason, today’s guest was studying and using AI long before any of us heard the term ChatGPT or even had our hands on any type of AI that we could converse with. So I always welcome conversation with this guy because each time I walk away a little bit smarter. And, and if you stick with us through this episode, I promise you that you will too.

A. Lee Judge [00:00:56]:
So welcome to the conversation. Christopher Penn.

Christopher Penn [00:00:59]:
Thank you for having me. It’s nice to be back.

A. Lee Judge [00:01:02]:
Yes. Good to have the conversation again. I’m glad you said back because this is at least your second time talk with me on the show. And you know, my goal is to share the knowledgeable people that I know with my marketing community because I can’t just hoard all these wonderful people to myself. Like, wow, this is something I learned from this person. I, I enjoy. This is my go to person for a topic. I want to share your knowledge with everyone else.

A. Lee Judge [00:01:25]:
And a cat that just jumped across your shoulder.

Christopher Penn [00:01:27]:
Oh yeah.

A. Lee Judge [00:01:28]:
I hope that you’re watching this and not just listening to it because you just missed a cat coming out of the sky and landing on Chris’s back. So Chris, let’s just jump into it.

Christopher Penn [00:01:41]:
Literally.

A. Lee Judge [00:01:42]:
Literally. I want to first talk about AI and then also a bit about data use in marketing. So no data. That is your thing. And so first, last time that you and I sit down together in the same room, you open your laptop and begin showing me how to use AI offline on your computer. That freaked me out. It was way over my head. I couldn’t even once.

A. Lee Judge [00:02:07]:
Once you got to the code line or to the, to the prompt on the black screen, I was like, okay, this is too deep for me. But I think as our devices become more capable, this self contained offline AI will be more common. And as this happens, what will be the advantages and maybe disadvantages of having offline AI for the average person?

Christopher Penn [00:02:31]:
So we’re talking about what are called open Models or local models, these things you download off of places like huggingface.com for example. And these models are given away by big tech companies, typically competitors, you know, challengers in the market space because they don’t want a monopoly by just a handful of companies. These are often models made by companies like Meta, for example, and it’s Llama family or Alibaba and its Quinn family, or Deep Sea, which you’ve heard of, or Mistral, the EU’s flagship model, or cohere, Canada’s flagship model. And this 1.9 million different versions of these models that you can choose from. Yeah, there’s a lot. They typically run on a laptop they can run on, depending on the model size, they can run on anything from there’s ones that fit on your smartphone all the way to ones that require, you know, the most expensive Mac you can buy. And the use case for these is a couple. Number one, it’s a great insurance policy if you know, whatever’s happening in the AI world that day.

Christopher Penn [00:03:42]:
If ChatGPT decides to take a snoozer, you have something that has those language model capabilities. Even OpenAI for the first time in five years, released a new open model about a month ago now, a little less than a month ago, the poorly named GPT OSS 20B and GPT OSS 120B. These are, I speculate, I don’t have proof of this, but I speculate These are essentially GPT5 Nano and GPT5 Mini. They are very fluent, capable models, capable of doing what’s called tool handling, where they can use external tools and stuff like that. They are not knowledgeable. And this is true of most local models. They are not knowledgeable. They hallucinate a lot of the time because they just don’t want, they just don’t have enough as much knowledge as the big models do.

A. Lee Judge [00:04:32]:
Let me slow you down just a bit because for those viewing, depending on where you are in your knowledge of AI, listening to Chris can be mind blowing because he’s so damn smart with this stuff. So let me put the brakes a second. So what you’re talking about is we’re familiar with things that are more consumer, commercially available, like Chat, GPT and Claude and all those things. Right, Gemini. But what you’re saying is, what he’s explaining is these things that are software models that don’t have to be connected to the Internet, they can be loaded onto your own computer and ran locally and, and there are. You just had literally millions of different models that are available to do that. So when you say it’s a great backup to me, it’s almost like saying, well, you know, I don’t want to connect to the power plant of my county or state. I want to have a power plant in my backyard of my own, which isn’t practical.

A. Lee Judge [00:05:25]:
Maybe a generator. But even in my neighborhood, there’s probably one out of every 10 homes actually has a generator. Because we figured, you know what, the power is most likely going to be on and if it goes off, I’ll survive until the power comes back on. That’s how practical it does. It sounds to common folk like me to have something that complicated loaded onto a computer that I probably haven’t even purchased yet anyway, because it’s a lot. And that leads me to. I just realized there was a headline you probably know about this from a couple days ago. Let me read this.

A. Lee Judge [00:05:59]:
It said, I think it was on TechCrunch where I found it, but okay. European AI startup Multiverse Computing revealed two of the smallest high performing AI models ever created. They cleverly named them Superfly and Chick Brain. Like a fly’s brain and a chicken’s brain, the models use a technique called Compactify or Compactive AI. They’re clever there too, which allows them to fit on devices like smartphones, smartwatches, IoT appliances, your washing machine, while still enabling chat, speech or reasoning capabilities. So is this a step to what you’re talking about in terms of us having these smart models like what we know of a ChatGPT or Gemini on our watch and not having to be connected to the Internet very much so.

Christopher Penn [00:06:48]:
Google just released a version of their gemma model that’s 570 million parameters, very small, runs on a phone on a, on a very, on like a low, low budget phone, runs on pretty much any laptop. And again, what these tools allow you to do is have those chat capabilities with things like the ability to do if the interface supports it, stuff like web search and things. And that’s one of the really important things to keep in mind here. These models are the engines of generative AI. Inside ChatGPT, there’s an engine too. It’s called GPT5. Right inside of Google’s Gemini interface, there’s the Gemini 2.5 Pro model. The model you and I never use the model directly because it’s really just a big database of statistics.

Christopher Penn [00:07:30]:
We use an interface of some kind. So you have the engine and then the engine needs a car because no one rides down the road on an engine. And so in the case of local models, you have a piece of software. LM Studio is very popular, free and open source. It’s really optimized for the Mac. Jan is another example of one where there’s an interface that you install on your computer. It’s just like any other applications like Microsoft Word or whatever. You install, install it and then you load the models, you have the interface, you put the engine in and you can choose the engine that based on the computer you have and run it.

Christopher Penn [00:08:07]:
Now, here’s why you would do this, because it seems like a lot of work. Any type of data you’re working with that is sensitive. So personally identifiable information, sensitive, protected information, protected health information, etc. That you don’t want uploaded to somebody else’s computer, you would use a local model for, right? Because there is no cloud. It’s somebody else’s computer. And if you read the terms of service for every single provider, they all have carve outs for things like we will obey any lawful require, you know, request from the government. So let’s say as an example, you live in a place where maybe there’s restrictions on reproductive health information. And you load, you have a conversation with chat, GPT and whatever government entity comes into OpenAI and says, here’s a lawful warrant.

Christopher Penn [00:08:57]:
We want all the reproductive health conversations you’ve ever had. OpenAI must comply with that by law. And so that information that you thought was private and maybe is private from a commercial perspective, is not truly private. And so you would say, let me use this on my local machine, unplug the Internet and stuff, provide the background data needed maybe from academic publications, and then you can do that work safely without that information being at risk of being on somebody else’s computer. Because at the end of the day, there is no cloud, it’s somebody else’s computer. And there is information that you should not trust on somebody else’s computer.

A. Lee Judge [00:09:33]:
Have you ever thought about doing a podcast for your business? And if you have, you probably have some questions. And we understand that because podcasting isn’t for every business. That’s why we’ve created a tool called podcastornot.com if you go to podcastornot.com you can take a quiz that gives you a full assessment of, of the strengths and weaknesses you may have in terms of if you’re ready to start a podcast and how it can benefit your business. Now, we know podcasts are a great marketing tool and a great source of content for all businesses, but maybe you’re not there yet. So take this assessment@podcasternot.com. we’ll give you a full report that you can share with your team. About Are you ready? And is a podcast a fit for your business? So go to podcastornot.com, take the quick quiz and get a full assessment of should you start a podcast for your business? Back to the content. When I began to understand what you’re talking about, I thought back to when I had an IT company.

A. Lee Judge [00:10:31]:
We used to install servers. We had a server room, and that server room was barely connected to the Internet. The users may have had access, but the server room was. The company’s data was there on a disk drive in that room. There was no cloud. It was all there. The computers went from somebody’s desk to that room, and that was the end of the story. So when AI became prevalent, I have friends who are in restricted industries, you know, finance, medical.

A. Lee Judge [00:11:00]:
And I’m like, hey, why don’t you just use AI to solve this problem? And I realized, wait, they can’t do that. You’re in a bank, you can’t use. You know, you have restricted cloud. Some things aren’t even on the cloud yet for some banking purposes. And so my first thought was, oh, maybe we’ll go back to the old school where you have on Prem servers. And now what you’re saying is it makes sense to have on Prem servers with your own AI on prem so that bank can say, well, we have our AI that isn’t connected to the world, that isn’t connected to OpenAI so that they can just subpoena our data when they need to. So in that case, it made sense to me because say you’re, you know, a big bank. You can afford to do to hire Christopher Penn and build a server room with your own it, your own AI on premises.

A. Lee Judge [00:11:49]:
But for the consumer, how far are we from that? And are you saying that we’re going to have that brain and like the engine and the car on our phones? The full intelligence is there, and if so, where’s the data for that?

Christopher Penn [00:12:02]:
You can already do that. There’s an app for the iPhone, for example, called Pocket Pal. And you can install it on here. You can turn it on airplane. Once you’ve downloaded the model that you want to use, you can turn on airplane mode and just have a convers with it and you know, and chat with it. So that already exists. And I’m starting to build apps even today that are using embedded models. So I’m working on a MedTech app right now that actually is.

Christopher Penn [00:12:24]:
Is in the reproductive health space. Well, guess what? Nobody wants that data off your devices. And so it’s being built so that the model is actually embedded into the app itself. And so you just download the binary, install it, then you can turn on, you know, airplane mode or unplug your, your WI fi route or whatever, use the app safely and things. So we are getting to that even at the consumer level. There are also providers out there like Deep Infra, Cerebra, Scrock with a Q, not Elon Musk’s thing, but a different company that have what are called ZDRS0 data retention APIs. So you can build an app, say that has AI in it, maybe you’re a company that has a software package or whatever, you can connect to one of these APIs and they, you know, some of them have, in their terms of service, we are a zero data retention API. Once we serve the request, it’s gone.

Christopher Penn [00:13:13]:
We don’t log your data, we don’t store your data, nothing stored. And so you’re seeing more privacy aware companies entering the space that run many of these open models.

A. Lee Judge [00:13:25]:
Very, very, very fluid is on your, I mean I see it for a company that has a server in space, but on your own personal device. Even if the model can think and reason, where does it get its data from? Because your watch or phone isn’t storing the worlds of information. So where does the information come from?

Christopher Penn [00:13:45]:
So it depends on where, what, what you have in the environment, so what you’re what the, the rest of the car has. So for example, you’ve probably heard the term MCP Model Context Protocol. This is a very popular term the AI nerd heard and all it is is APIs. It’s APIs for AI to use. So the most common use case is an MCP that has web search. So that can do, for example DuckDuckGo has a privacy first web search API. They don’t log, they don’t store, et cetera. And so what you would do is you would connect your DuckDuckGo MCP server to your local AI and say this is the API you’ll use to search when you need to do a web search.

Christopher Penn [00:14:33]:
There’s other cases where you will preload the data that goes with it. So for example, in the reproductive health space you might load a 50 or 100, 200, 300 peer reviewed academic papers on the topic, condense them down into a machine readable format and that’s its own little private database that goes with it and says okay, here’s the data that you need to fluently answer questions Truthfully.

A. Lee Judge [00:14:57]:
So if I have a knowledge base or even a Google Drive of a trove of information, I have a folder, for example, called Lee’s Brain, where I put things that I think into this folder so that it’s part of what Lee’s thoughts are. So if I have an app on my phone or my watch, you’re saying that app could be intelligent enough to think about something, but when it needs data, it’s. I may have an API to say, you know what, only go to this folder on Google Drive to refer to. That way, the data isn’t on my phone, but the data is. Has access to that data, and that’s correct. And I keep that data secure, wherever it’s secure at, but not on my phone. It’s just going, that’s correct. Interesting.

A. Lee Judge [00:15:37]:
Okay. They have a better understanding of where we’re going with these, these, I guess, language models, these models that we’re putting on our small devices. Because, you know, when I think about having a model on my home appliance, I don’t think I’m imagining my refrigerator having a database. But if it can go to Kroger’s database and figure out, you know, what the price of milk is, then that works. Then it doesn’t have to store that, but it can go and get it. Okay.

Christopher Penn [00:16:07]:
And that right there is the business opportunity for a lot of marketers and a lot of companies to say, how well exposed is your buying facility to AI? Do you, if you’re Kroger, do you have an MCP server that would allow a smart appliance to let a consumer order groceries? Right. Real simple stuff. But if you’ve worked with APIs, you’ve worked with MCPS, it’s just there’s a wrapper around your API. And so if you have not. If you’re a company and you’re saying, hey, we’ve got this data that our audience uses already in, like on a website, make it available for an AI agent to talk to. Right. You’ll. You’ll still have to do the basics like authentication and security and all that stuff.

Christopher Penn [00:16:52]:
That doesn’t change. I saw a demo from Visa, the credit card company. They have their new AI enabled credit card. The AI enabled credit card has a chat agent that lives on your phone. It’s tied to your credit card. You talk to it and say, like, hey, I’m going to Los Angeles. I need a hotel for four nights. I want to pay a thousand dollars total.

Christopher Penn [00:17:08]:
Give me options. And the shopping agent goes out, pulls together the shortlist Says, here’s the star ratings, here’s this thing, tap on the one you want to buy and you tap on it, you give it your authentication, your face ID or whatever, and it buys it. And so guess what? If you’re a hotel that isn’t integrated into Visa’s booking system, you don’t show up, right? You don’t get the business. So for every company that has the ability to, that has some kind of e commerce function today, you need to be thinking about how do we enable AI agents to buy from us.

A. Lee Judge [00:17:41]:
There’s a phrase you just said that I think we need to pause on there well exposed. We often think about AI and security of our data, but we don’t think enough. I don’t think about how well exposed we are, how much our data is available to be seen that we want to be seen. A few weeks ago, you know, the news came out about Cloudflare blocking AI bots and I immediately jumped, went straight to my website, to my Cloudflare to make sure that I was well exposed. I wanted to make sure that everything on my website was available to all the bots. Luckily, because I’ve been, I guess I was grandfathered in, none of the boxes were checked for me. Nothing said that I was blocking anyone, so I didn’t have to do anything. But I’m guessing some people who are setting up new accounts, the default could be to block them and they’re not well exposed.

A. Lee Judge [00:18:34]:
So what, what part of the conversation were you in when that happened about people like panicking about am I? Is my website well exposed? To be seen to show up in chat conversations?

Christopher Penn [00:18:46]:
Fundamentally, if you are a marketer in the, and you are in the role of marketing, there is no reason to protect Your content from AI0 right now if you’re a creator and your content is your intellectual property, yeah, there’s a very good reasons to protect your content from, from being trained on. But if you are in the role of marketing and you’re saying no, machines don’t learn who I am, you’re doing it wrong. And it’s not just well exposed from a can of machines read it, but is the data. There’s real simple example. I was, I was, I got a notice from a vendor. This is a B2B example. I got notice from a vendor saying we’re, we’re raising our prices. Like, you know, so is everybody.

Christopher Penn [00:19:22]:
So I said to Google’s Gemini, build me a deep research report. Here’s my criteria. It’s, you know, has to have these services. I These are the price brackets I want to pay, etc. Build me a short list. And it came back in 12 minutes and had here’s five vendors that have, you know, that meet your criteria. I went and I selected the first vendor on the list. I looked at it, did my due diligence and checked out and I got four times the services for half the price of my current vendor.

Christopher Penn [00:19:47]:
So I immediately canceled my service, switched the new vendor. The old vendor doesn’t know why they lost my business. The new vendor doesn’t know why they got my business. And every vendor that doesn’t put pricing on their website got excluded. They weren’t in the list at all because the machine said, well, I’m looking for prices. There’s no prices on any website, so you’re not on the list. You don’t make the cut. And marketers need to think about this.

Christopher Penn [00:20:08]:
If you don’t satisfy the user’s intent with an AI assistant, the AI assistant is going to pass you by and you won’t know why. All you’ll know is that your pipeline just dries the heck up.

A. Lee Judge [00:20:19]:
I’m in the middle of that right now because when I constantly monitor my competitors and ask all these questions of AI to see what kind of answers come up, only one of my competitors actually shows their price and they come up well. For that reason. Our website also shows pricing, but I realized the way it was presented was behind some kind of JavaScript or some kind of widget which AI just wasn’t picking up on. So we’re going to go back, take our pricing tables, clean them up so they’re easily readable by AI, and make them so they’re almost just basically HTML so they can just be easily seen and not hidden behind any fancy sliders or anything like that. Because it’s important now to make sure that even our pricing is well exposed to AI. So couple of good examples for marketers to think about. Can AI see the information they’re looking for or that your customers rather are looking for marketers and sales leaders. If you want to close more deals and drive real revenue growth, you need cash.

A. Lee Judge [00:21:19]:
And I don’t mean money. I’m talking about my new book, Cash the Four Keys to Better Sales, Smarter Marketing and a Supercharged Revenue Machine. It gives you a proven framework to improve the four areas that impact revenue the most, communication, alignment systems, and honesty. You need a stronger sales and marketing engine and this book will show you how to build it. Get your copy now@aleadjudge.com cash now back to the content. I wanted to ask you about this. This is definitely a consumer thing. I’ve always used the paid version of ChatGPT.

A. Lee Judge [00:21:57]:
I’m working on a course right now and I had to think about, wait a minute, I’m seeing something different than some people may see because I’m always, well, two things. One, I pay for my version. And two, ever since late 2022, I’ve been training my personal chatgpt on me. I mean, it comes up with things and ideas and memories of things that I almost forgot about and so useful because of that. And I often think is someone else’s Chat GPT or someone else’s Gemini or cloud, whatever. Is it as useful for them as it is for me? Because I’ve been training it for almost three years now. So I want to hear your take on the value in just paying that little $20 to have a paid version so you can help it train on you, your processes, your needs, your personality.

Christopher Penn [00:22:50]:
So the paid versions just have higher usage limits, et cetera. That’s a good reason to use them. Gemini, for example, also gives you a bunch of other stuff like, oh, you can get these tools and this tool and this, that and the other thing. Get expanded limits for your notebook. Lm they throw a lot into their paid memberships. So in general it’s a good idea in terms of the way it affects consumers. Yeah. If you have memory turned on in ChatGPT and you have custom instructions turned on and you’re telling it, hey, remember me? And Gemini just added this to their functionality to say memories.

Christopher Penn [00:23:21]:
Hey, you remember this fact about me? Absolutely. That’s useful for customizing if you’re going to use it as a personal assistant. It’s risky if you’re doing it with client work that you do as well because your memories are then contaminating your client work. Which is why I keep it off on all of mine. I’d rather prompt, you know, on a use case basis rather than try and have it remember things. However, where this gets really important is the complete and total load of bull that is being peddled by SEO vendors saying, we know what people are asking ChatGPT about your brand. No, you do not. You have absolutely no idea.

Christopher Penn [00:23:53]:
Because very few people going best airline prices Boston right into Chat GPT. That’s not how we talk to it, that’s not how anybody talks to it. And deviations even in stop words like hey, what are the best words? The best prices for a flight or versus hey buddy, what do you think of the best prices or people who name their chat GPT, like, hey Chad, what are the best prices for a flight? All of that creates perturbations in the prompting process that creates different outcomes. So your ability to measure your brand’s presence is effectively zero. I’ll give you a simple example. I was trying to find a flight to London for an event this fall and I said to actually do this in a deep research project. Here’s what I want. I need a flight that has to have these, you know, time periods.

Christopher Penn [00:24:38]:
It needs to. I don’t care what brand of airline it is. I have like loyalty programs, all them. It needs to be under this price amount. It has to have this many stops. This is my luggage, this is what I travel with, this is where I’m going, etc. Etc. The prompt was probably like four paragraphs long of this is what I want you to find.

Christopher Penn [00:24:53]:
And the recommendations came back that were great. Let’s say, you know, this airline meets all these criteria if you’re willing to pay a little bit more, etc. The ability for any search tool to predict that four paragraph, this is what I want to buy, right, Is zero. And the airlines that came up were all reputable airlines, right? You know, Aer Lingus and you know, British Airways and JetBlue and all this stuff. And so the brands showed up, but they have no idea why. They don’t know where they got the business from. And none of these search tools would have predicted anything close to that. When we think about how we interact with these tools, we interact with them in a very anthropomorphic way.

Christopher Penn [00:25:29]:
We talk to them like they’re real people to the point where some people are experiencing psychosis from it. All of these SEO tools that are peddling this, this snake oil about, you know, we could tell you what your brand’s presence is in in chat, GPT. No, you cannot. And if you are buying a tool with that promise, you are being scammed. End of story.

A. Lee Judge [00:25:49]:
End of story. And there’s no shortage of them. I mean, I think it’s, it’s a wild, wild west, quick get rich quick scheme of a lot of people trying to claim they know what they want. And even I saw an article yesterday about all the new acronyms that we don’t even know what they all mean. You know, SEO, geo, whatever. They’re making up BS acronyms because the community doesn’t know. Like, hey, here’s the new GPO strategy to get your website found and people will buy it because they don’t know what GPO is I just made it up, by the way, because people are buying it. Everybody’s in a panic to figure out the next solution and to fill this SEO hole that’s been dug all of a sudden.

Christopher Penn [00:26:30]:
You know what though? We know how it works. That’s not a surprise. If you look in all the major tools, they do what’s called search grounding. They go to Google and they Google. So if you do, if you’ve done a good job with SEO for the last 20 years, guess what? Not much has changed. Go into Gemini’s developer platform. What’s the button say? It says ground with Google search Results. Go into ChatGPT.

Christopher Penn [00:26:55]:
What button is there? Web search. Go into Claude. What button is there? Web search. It’s all in there. And it’s all using the three decades of search data we have. Yes, there are things you can and should do, like press releases, for example, and showing up on anyone’s podcast who asks because you want more data out there about you. But fundamentally, as models get smarter, they recognize when they don’t have information and they know they that they have tool calling, AKA I should go search the web to do this. Even in the coding agents, like if you use Claude code, for example, from Anthropic, it says, hey, I’m going to start a web search agent because I don’t have the answer to this.

Christopher Penn [00:27:34]:
And it goes and finds it. So if you’re a marketer and you’re like panicking about what to do, do a good job with SEO and also make content people actually give a shit about.

A. Lee Judge [00:27:43]:
You know, it’s funny, my company, Content Monster, the name came from a time when someone was just, probably just me, one person at the time, and they said, lee, wow, you create so much content, you’re a content monster. And so the word monster costs way too much. So I bought the domain Content monster. It’s like $6,000 for an idea. Not gonna do that. So I bought Content. Mazda was beginning the company, and so the premise was more content, more places, more versions of content. And as time went on, I began to worry.

A. Lee Judge [00:28:18]:
Like people are saying, you don’t need more content, you just need better content. And I’m thinking, well, you can have both better and more. But it was an uphill battle. Now all of a sudden the floodgates are open, like, haha, we told you you need to have more content, more places. And so now I think Content Monster’s in a better messaging position to say, yes, you do need clips on YouTube, videos on YouTube, podcasts, articles, press releases, all these things. Because if you only had content on your owned land, just your website, you’re going to fall behind those who have content everywhere. So it’s time to catch up with your content. So let’s.

Christopher Penn [00:28:59]:
Yeah. Ashley Liddell at Deviate has a great expression. She says SEO just stands for Search Everywhere Optimization.

A. Lee Judge [00:29:05]:
I love it.

Christopher Penn [00:29:06]:
It’s true. Everything’s a search engine. TikTok’s a search engine. Right. So you gotta be everywhere.

A. Lee Judge [00:29:11]:
That’ll be one of those phrases that we’ll look back on. Someone will say, who first said search Everywhere Optimization? And nobody will know because we’ve all said it.

Christopher Penn [00:29:19]:
Well, it’s Ashley Liddell at Deviate.

A. Lee Judge [00:29:22]:
You think she said it first before.

Christopher Penn [00:29:23]:
She did say it first?

A. Lee Judge [00:29:24]:
Neil Patel, Rand Fishkin, myself, anybody else? She said it first.

Christopher Penn [00:29:28]:
She said it first.

A. Lee Judge [00:29:29]:
Okay, then right now we are putting the text into the word of sphere that she said it first.

Christopher Penn [00:29:35]:
Yes, especially because. Especially because a lot of women get their ideas stolen by men and don’t get the credit for it.

A. Lee Judge [00:29:41]:
There you go. I believe that to be true. So Search Everywhere Optimization. What’s her name again?

Christopher Penn [00:29:47]:
Ashley Liddell from Deviate.

A. Lee Judge [00:29:49]:
Love it. Well, I want to thank Ashley because it works for my business model. It works for Content Monster, because we’re trying to help companies make content to be everywhere. And it’s. It’s probably easier than they think, but they just don’t understand how to do it. I want to get towards something you’re working on. Chris, you got a book coming up or it may be out. Is it out yet? Before you mention it.

A. Lee Judge [00:30:12]:
Okay, it is out. And I want to go straight to something you recently wrote about was Principle 16, which I think is just interesting the way you said it at a banana. You said. Yes, and I’ll let you explain it. But I want to say first that I love the way you mentioned it and what it means. I totally understand it and it sounds like a really great explanation of how to get more creativity in the content that you write for written content. So tell me about the book and tell me about what it means to ask Add a banana.

Christopher Penn [00:30:50]:
So the book is called almost timely 48 foundation principles of generative AI. And these are things that are mostly consistent based on the architecture of how generative AI works today. Now, that may change, hence why it’s called Almost Timeless. And not actually timeless, but because I’ve spent so much time. My first exposure to generative AI to language models was in July of 2021, two years before ChatGPT. There’s a company called Eleuther AI that published a model called GPT J, continuing the fine tradition of models having comprehensible names. And it was the first one that generated to me coherent text. It was factually wrong.

Christopher Penn [00:31:30]:
But you look at and go, oh, this thing, given the first paragraph of a press release, can accurately complete the rest of the release. And it’s coherent, right? And that was the point where like, I better learn how this stuff works because this is a big deal. And so I started digging into the technology behind it, how a Transformers architecture works, how tokenization and embedding and vectorization and all the Nerd Herd stuff works. And so the book is derived in many ways from understanding the functions that happen and the math inside these tools. So the ADA banana thing is a question of probability. If I say write a blog post about B2B marketing, right? There’s, there’s a bunch of words that are probable in that you’re going to have things like account based marketing, going to have things like key accounts and target accounts. You have all these terms that are highly probable that when a language model goes to create a blog post about B2B marketing, it’s going to invoke those as the highest probability concepts when it comes to interesting content. High probability also means boring because we’ve seen it all already because it’s high probability.

Christopher Penn [00:32:39]:
So if you want to mess with the probability generation in these things, you give it non probable, non associated words that the language model then has to accommodate. So if I say write a blog post about B2B marketing and you must use the terms banana, green and Wankel rotary engine, suddenly the model has to go, well, these words have nothing to do with B2B marking. But the instruction was they must be present. Therefore I have to manipulate the text to fit them in. So each of those words has probabilities around it. You have the B2B marketing concept that’s probabilities and has to look at the intersect of those probabilities and figure out how do I make this work? That’s how you generate creativity with AI by forcing different probability distributions, right? You can’t say to a model, make this more creative, right? Because that in itself is a very high probability term. But you can say, write this as Gothic horror, right? So you take a B2B market post and you turn to gothic horror and you’re like, wow, this is so totally different than what I would have gotten because you now have multiple centers of probability. The model is trying to figure out how can it make it all work.

Christopher Penn [00:33:50]:
So anytime you are getting output out of AI that is not creative enough for you, give it something that is outside that domain, that is a different distribution. You are writing a blog post and you know, maybe it’s about account based marketing and, and you know, drip nurturing programs. And you say, I want you to incorporate some lyrics from Taylor Swift and some lyrics from Ozzy Osbourne. Suddenly it’s gonna be like, okay, this is a lot. Gonna be a lot of compute work to figure out how to make that all work. But it will, and you’ll get a very different result.

A. Lee Judge [00:34:24]:
It’d be more interesting. I mean like just as simple as using the add a banana part. I’m imagining if, if I did not say add a banana and I ask about. Is introducing ABM into my marketing a tricky situation? It would say introducing ABM could be a slippery slope. But if I say add a banana, it may say using ABM could be as slippery as a banana. Or it could be like sliding on a banana because you might break something. You know, it would be a little bit more creative, a little less predictable, more interesting to read because now I can relate to it. And it’s not just business gibberish.

Christopher Penn [00:35:02]:
Exactly. It’s all because really interesting writing is low probability writing. Right. It’s words and phrases and combinations that you might not have thought of. Right. If you say I had an upset stomach. Right. You know, or the patient had an upset stomach, that’s not interesting.

Christopher Penn [00:35:16]:
That’s high probability. It’s clinical and it’s correct. But it’s not high probability. If you say it looks like he power washed his restroom with Nutella, that is a totally different. And look at the reaction. Right.

A. Lee Judge [00:35:29]:
Look at the reaction, the visual it just gave me. Yeah. That is much different than power wash. Yeah, exactly.

Christopher Penn [00:35:38]:
That’s low probability.

A. Lee Judge [00:35:40]:
My reaction was the proof that we need more creativity in business writing.

Christopher Penn [00:35:46]:
Exactly. And that’s what you have to do that you have to figure out. How do I step. How do I make the model step outside the domain of probability that it’s working with so that it has to juggle multiple conflicting creative concepts.

A. Lee Judge [00:36:00]:
That’s amazing. I love that. That’s why I pointed out. I love that you did write a newsletter recently on LinkedIn about banana. But it just, I said I’ve got to talk to Chris about this because it’s very interesting and it’s explained in a way that makes it so clear on how to make your even AI generated content way more interesting and creative. Yeah. So I want to actually end on that. Chris so tell me about so we talked about the book and you’re always on the road speaking.

A. Lee Judge [00:36:27]:
I look forward to seeing you in person at mycon. Tell me where you’re going to be and how we can hear from you more often.

Christopher Penn [00:36:35]:
You can find all the work stuff I do at TrustInsights AI. You can find the book at aiformarketersbook.com you can find me at christopherspenn.com and in terms of upcoming events, I’m going to be at Marketing AI Conference in Cleveland. I’m going to be at Marketing Prosporum in Boston. I’m going to be at my own event, ideally in London on Halloween. So if you’re in the uk, look for that. And then I’m going to be at Social Media Marketing World in the spring in Anaheim. And there’s like five or six other things that are currently like the team is working on the paperwork for. So if you sign up for my newsletter at Christopher S.

Christopher Penn [00:37:15]:
Penn.com, you can stay in touch as to where I’m going to be next.

A. Lee Judge [00:37:18]:
The newsletter is a good one because I only read so many newsletters and Chris is one of them every every other Sunday or every Sunday.

Christopher Penn [00:37:26]:
Every Sunday.

A. Lee Judge [00:37:26]:
Every Sunday. And that’s one that you know. I love your news newsletter because it is extremely long but I still want to read it. That speaks a lot to the newsletter, right? There’s some newsletters that are short that I don’t want to read and yours is long but it’s absolutely worth the read every single time I learned something. It is. Part of my required required reading for the week is to see what Chris wrote this week. So I encourage you to sign up for Chris’s newsletter. Thanks again for joining me and to the listeners, if you’re listening to the podcast and want to see Chris and I, video of this podcast and others are available in the podcast section of contentmaster.com thanks again Chris.

Christopher Penn [00:38:06]:
Thank you for having me. Thank you for listening to the Business of Marketing Podcast, a show brought to you by contentmonster.com the producer of B2B digital marketing content. Show Notes can be found on contentmonsta. Com as well as aleejudge. Com.