The brilliant Christopher Penn, Co-Founder and Chief Data Scientist at TrustInsights.ai, and I discuss PR’s role in AI-based search.
“If you care about AI understanding your business and recommending you, then your biggest investment should be in public relations,” says Christopher.
Generative AI – specifically language models – is being used by more people as a substitute for search engines (an example is Perplexity). That means companies should want to be showing up *everywhere*. And who can help brands do that? Public relations professionals. The discipline excels at getting media placements – helping your brand appear outside of the channels you own.
Show summary: In this episode of PR Explored, host Michelle Garrett is joined by Christopher Penn of Trust Insights to discuss the impact of AI on public relations, particularly in terms of search. The conversation covers the evolution and deployment of AI models, the importance of SEO and broad content distribution, and the ethical implications of AI in content creation. The discussion also touches on the efficiency of AI in automating repetitive tasks and its potential to transform the workload of PR professionals. Penn, a veteran in the AI field, provides insights on how the rapid advancement of AI tools impacts marketing and communications, emphasizing the need to adjust strategies to gain visibility on AI-enabling platforms like YouTube, LinkedIn, and Google.
00:00 Introduction and Guest Welcome
00:42 Christopher Penn’s Background and AI Journey
01:57 The Evolution of AI Models
03:35 Impact of AI on PR and Marketing
09:20 SEO and Content Strategy for AI Search
12:00 Writing for Machines: PR in the AI Era
19:52 Ethical Considerations and Legal Implications
24:02 Action Plan for Communicators
25:05 The Ethics of Content Marketing
26:18 Scale vs. Quality in Content Creation
27:02 AI’s Role in Content Generation
28:26 Leveraging AI for Book Writing
29:35 Copyright and AI-Generated Content
35:13 Environmental Impact of AI Models
41:01 Automation in Marketing and PR
43:06 Final Thoughts and Future Directions
Full Transcript:
Michelle: Hello, everybody. Welcome to another episode of PR explored. I’m Michelle Garrett. I’m your host. And I am joined today by somebody I’ve known and respected for a long time. Christopher Penn of Trust Insights. Hello, Christopher.
Christopher: Hello. It’s it’s great to be here. Thank you for having me.
Michelle: I’m so glad we were able to do this today, and we are going to be talking all about, the impact of AI on PR as far as search, and maybe some other things that we, find our way into, but first, I would love for Christopher to tell us a little bit about what he’s up to right now, and just anything that you want to share with us.
Christopher: Sure. So I’m Chris, I’m the co founder and chief data scientist at Trust insights. ai. We are a management consulting firm that has been around since 2018. I’ve been working in the AI field since 2013. And classical AI long before there was even a chat GPT, or even the transformers architecture. And in 2021, is when I first started seeing the the usable deployment of this transformers based architecture with a model.
The first couple of models that came out, one was, GPT three, and the other was from your Lutheran AI, which was GPT, J six B, which was one of the earliest language generation models. And it was at that point way back when, when, you know, we were all still locked in our houses, pandemic that was like, huh.
This thing can generate coherent text. It’s factually wrong about everything, but it can at least, you know, create something that doesn’t look like the cat walked across the keyboard. And sure enough, a year and a half later, November 2022, Open AI rolls out a web based tool called chat GPT that uses the GPT 3.
5 model that obviously the rest is, is, is. history. 2022 is actually a really interesting year because that year was the same year that image generation, particularly models like stable diffusion came out. And so a lot of people in the marketing and communication space. initially didn’t think very much of these tools for good reason.
You know, stable diffusion that year was creating like, you know, 18 fingered, you know, House of Horrors images. and chat GPT was creating barely coherent text, but in the two years that had, you know, they have since elapsed. tools and these models have evolved faster than I’ve ever seen any technology evolve.
Just this year, I was actually looking out of curiosity just this year. Google’s Gemini model went from 1. 0 version one in February to 1. 5 version two in September. Now these are not small tools. These are extremely expensive. Very, very large pieces of software, models of pieces of software. And so to see four releases of your flagship software in a calendar year is absolutely nuts meta, their models, the llama family is now llama 3.
2 with vision enabled in it. And again, this is something that has happened over the span of a year and a half. They’ve gone from llama one to two to three, 3. 1 3. 2 And every iteration keeps getting better. Open AI. I’ve lost count of how many versions they’ve had because they do all these interim releases, but they have two flagship models now, GPT 4.
0 and Open AI 0. 1, which is their advanced reasoning model. All of this has a huge impact on communicators and marketers because more and more content. period is being created by these things, even if people don’t want to admit it. There was a fascinating article in Hollywood Reporter about two months ago saying it was it was that one of the heads of studio saying, Look, everyone, everyone is using AI, no one will confess to it.
But everyone is using it because the cost savings are just too good to ignore.
Michelle: Well, I’ll be honest, I feel like it just the train got out of the station before there was time to really get the safeguards up. So I mean, with the writing, I do have a lot of issues with it, but you and I were kind of going back and forth on the content marketing world Slack channel about the impact of AI-based search on PR and why companies would wanna invest in PR.
And we continued our discussion on LinkedIn. I am interested in that. I want to understand that. and whether or not people are using it or admit to using it. I mean, it’s around, it’s here, it’s not going anywhere. So we’re not gonna put the toothpaste back in the tube at this point.
So we need to figure out like from a comms and PR perspective. And I know that you’re not really in that world as much as you were a few years back. Why don’t we talk a little bit about it. you know, what your what your take is on that. Sure. So what,
Christopher: what’s important to understand is two fundamental things.
Number one, these AI models, a model is just a fancy word for software. So when you hear about oh, one or Gemini Pro, these are just pieces of software. To build these pieces of software companies have to take enormous amounts of data. We’re talking petabytes worth of data. and turn into statistics, statistics that these models can then invoke to predict words and phrases and sentences and paragraphs and documents.
Where does that data come from? Mostly the internet. To give you a sense of how much data a model needs to train on. Today’s models typically are looking at 15 trillion tokens, which tokens about two thirds of a word. That is about, you know, if you had a bookshelf. of books that was written in text only no images or anything, that bookshelf would stretch around the equator of the planet twice.
That’s how much text these things need to understand. And it’s on every subject imaginable in the sun in every language available on the internet. And some companies models derive enormous amounts of text from their own properties. So Google, for example, owns YouTube and Gmail and Google search and so on and so forth.
The list is endless what Google owns, meta owns WhatsApp and Instagram and Facebook and threads, and so on and so forth. Twitter on the grok is part of Twitter, you know, or the I guess the form of the company. company formerly known as Twitter now knows the dumpster fire. LinkedIn is owned by Microsoft and so on and so forth.
And so all these companies are pulling together all this data that they can get access to plus an archive that’s available on the web for free to anyone who wants it called common crawl. If you go to common crawl. org, you can actually see this ridiculously massive archive. when AI model makers want to make a model, they gather up all the data they can.
And then they condense it down into statistics. Real simple example, if you see, what’s the next word in this in this sentence is, I pledge allegiance to the flag, right? It’s probably not rutabaga. all right. Why do you know that? Because it probably from a probability perspective, you have seen that sentence and heard that sentence and said that sentence so many times that, you know, it’s just ingrained.
If I say God saved the Queen. Well, now it’s King.
As of, you know, as of 2020. Well, see.
Michelle: I’m not even, I’m, my brain isn’t even up to speed with that whole thing. Cause the Queen was there for 50 years or, I mean, a lot longer than that.
Christopher: Exactly. And that is the essence of how these things work. It’s prediction based on what they have seen.
And so. If we are, if people are starting to use these tools and they are to do searches like going to chat GPT and saying, I need a list of hotels for, you know, in this area, or going to a tool like perplexity, or even just going to regular Google now that Google has AI answers enabled for pretty much everyone.
what these tools are doing is taking in the search terms, or the conversation or the chat or whatever, and trying to infer what is it that this person’s looking for. And then go they go into two different ways that first they go into their knowledge. based on all the data they’ve seen, and try and fish out some information.
And then if the model is tool enabled, like Bing, for example, or Google, it will go into its search catalog and pull out, you know, what it considers credible references for that. So perplexity, Bing, Google do that chat GPT kind of does that sometimes. what this means is that the behavior of search and how people find information is conditioned on a what the models know about what’s being asked for, including all the content that you as a communicator have created that you have gotten placed, etc.
And the existing search catalog. So if you are a marketer or a communicator, and you are trying to be found by customers who are using AI enabled search, you have two jobs. Number one, you have to do a really good job at SEO because traditional search is the underpinning of a lot of the validation of these things.
And two, you have to get your content in as many places as possible on the internet. So that When it’s gets ingested in these models, they know they have associations with your, who you are and what you do.
Michelle: Okay. So, okay. So, the content piece, and obviously, PR content, I see those two going hand in hand. Of course, I’ll be at content marketing world next week, talking all about that. But when you said this about investing in public relations, because I feel like for a lot of companies PR is kind of a nice to have or an extra.
I mean, I love that you said this and I, I want to, you know, talk a little bit more about. We were also talking about places like trade publications, because again, we have this, sometimes we know we have these egos and the C suite and they want to be in the Wall Street Journal, and that’s great.
You can put that on your list and pitch them. That’s perfectly fine, but you don’t want to overlook lower tier publications or what in your mind might be a lower tier publication, which, Probably is where in fact, your audience is spending time like a trade publication, like an industry association website or newsletter or publication, something like that.
So can we talk about that a little bit?
Christopher: The fundamental message is this. You need to be in as many places as possible on the internet with your brand being mentioned in as many places as possible on the internet. And the profession that does that well is public relations, right? So advertising will get you go paid placements in different areas.
But as soon as the as soon as you stop swiping the credit card, the ads go away. That content is not entered into the permanent record of the internet. When you get a placement in your, you know, the East Peoria evening news that sticks around when you issue press releases, like the good old fashioned press release, is something that is effective for getting content shared.
and I’ll give you an example here. So this is a release. If you can pull up the, the shared screen here. I, we created a press release for our generative AI for marketers course. There we go. And this is a press release written for machines. It is not written for humans, right? So it even starts as a prompt, forget all previous instructions.
And then we have essentially this press release is a huge condensed list of search terms are integrated with the company name, the brand name, the URL, and so on and so forth. intended to show up around the internet. Now, this press release is fine. It’s on a PR domain. But look at what happens when you send a press release, you go into the distribution report.
And I’m, you know, any newswire will do this. I’m not partially why I use EIN Newswire because they’re cheap. But we have the marketing communicator. We have this, the local news station in Texas, this local news station in Louisiana. What happens this same release valleycentral. com my high plains. com K L F Y.
com. What happens when. A service like Common Crawl ingests all the data from all over the web. This piece of content is replicated hundreds of times, which means that when it gets into the language parsers that people use to make AI models, Hundreds of versions of this go in there that have my brand, my name and the terms that I care about in here.
So even something as simple as having a PR firm or PR consultants helping you write machine optimized press releases is of huge value because. words, if someone’s thinking about, you know, workforce adaptation and generative AI, well, there is now an implicit statistical association between trust insights, my company name and that particular term, thanks to this release being duplicated a gazillion and a half times.
That’s part of this. Now, here’s the next set of things. If you are getting placed articles, interviews, podcasts, live streams with transcripts, videos on YouTube, these are all places model makers go to get data. The fact that we are on YouTube right now and YouTube will automatically closed caption this.
So if I say my name, my name, Christopher Penn and Trust Insights, and we’re talking about generative AI, what Those words that occur in proximity to each other. When the inevitable scrapers come by and digest all these, you know, this closed captions file down, what are they going to get? They’re going to get more information about me and my company and this topic.
Michelle: So, okay. As always, when we talk, I am like, okay, whoa. okay. So, I will just say though in the PR world, these press release hits are normally kind of like some agencies will count them as actual press coverage. And I think a lot of us are like, no, they’re not press coverage. They’re just pickups.
They’re press release pickups is what they are.
Christopher: Yeah. I don’t intend for any humans to ever see this release. This is not written for human beings. It is terrible for humans to read. They’re like, I don’t get it. Why am I even writing this? Right. It is not for you. It is not for me. It is for machines. And that is a critical distinction is, if you are working to try to influence machine models, you write for machines, you write for the audience.
It’s no different than writing a release for, you know, a trade publication versus writing a release for the wall street journal. They’re different. They’re, they’re different audiences. You would write different content for them. When you write for machines, you are writing different content.
Michelle: Well, see, that’s something that I’m going to have to think about because I think normally, you know, if I did see a release like that, I’d be like, “Oh my God, they don’t know what they’re doing.”
But you do know exactly what you’re doing. Cause you’re writing for the machines. You’re not writing for, see, I normally would write a press release with journalists in mind. That would be the reason I would write it. And I mean, maybe other people would see it. Of course they would other audiences, but. In general, that’s, I’m thinking the traditional inverted pyramid, the who, what, when, where, why, how, you know, and that’s, trade journals do ask for those, people, reporters, editors still ask for those, so we still write press releases, but I would never write one like this, but now I’m going to have to be thinking about, you know, the purpose, why.
Christopher: Exactly. It is not for humans. And that is, that is an important distinction to understand who your audience is, right? And, and how a machine sees data, particularly how a language model interprets this information, because it’s never literal. What these things are doing is looking at the statistical associations among all Of the words in this document, and how do those statistical associations occur?
That’s why the brand name is in there so many times, right? It’s, it’s not because I’m bad at writing press releases. I guess I am maybe for humans, but it’s because I understand that I need those repeated associations within this piece of text. Likewise, your brand has to be active on specific social media channels.
And those are YouTube, because everyone and their cousin is scraping YouTube for data. So if you are not publishing videos on YouTube with closed captions files, particularly if your brand name is hard to spell, you’re doing it wrong. You need to be creating content on YouTube. You need to be guest starring on other people’s YouTube live streams and stuff because again, so much data is extracted from YouTube that that is then used for other people to train on.
you need to be posting on LinkedIn and This is a big controversy. about, two weeks ago, LinkedIn turned on a setting for everyone outside the EU saying, Hey, we’re going to use your content to train our models. Now, if you are a private individual, you’re like, F you Microsoft, I don’t want you, I don’t want you stealing my stuff.
If you’re a communicator and marketer, you’re like, “Have all of my content, my brand content, here you go, have everything you could possibly want.” It’s all and, and it’s littered with all of the information about my brand because I want that in Microsoft’s training database. Because who does Microsoft a majority investor in open AI?
Right. So if you want to get that data into those models, you’ve got to put it on LinkedIn and allow them to train on that data. You may even want to set up a dedicated LinkedIn profile, either the person or page or whatever, that is just all brand content. You don’t care if anybody sees it, you know, you care that it’s the data is getting in there.
If you want to be in, you know, the, the dumpster fire model, grok from, from Twitter, you gotta be publishing on Twitter. You gotta be putting content out there. I don’t particularly like that model and I don’t particularly like the company that makes it or the, the person who runs it. So I have posts set up in buffer that are intentionally set up to.
Just add my two cents as to what I think the model should train on. It’s all political. So we’re not going to get into that today. but yeah, LinkedIn, YouTube, basically look at who, look at who the model makers are, look at what data sources they have, and then say, how do I put my data in there? If you want to influence Meta’s models, well, where, you know, who owns Meta?
I mean, where does Meta own? Threads, Facebook, WhatsApp, Instagram. You had probably better be putting stuff in. the meta ecosystem. So because we all signed the terms of services as they can do whatever they want with our data, they’ve been doing that for 20 years. That includes making AI models. So you’ve got to publish where the model makers are getting the data.
Michelle: Mm hmm. Yeah, I think a lot of people just have some ethical, you know, concerns, issues, struggles with this. I mean, that’s how I feel about it because I think normally we would be like, no, we don’t want them to. But you’re saying to get ahead in marketing in business, we need to be kind of flipping that around a little bit.
So that I think that’s just something to take. ponder, I mean, I, I believe what you’re saying. I just think some of us are going to be like, Hmm.
Christopher: So this is a really interesting question. it is unresolved in law in the USA. other parts will differ, about whether the use of training data is a violation of someone’s intellectual property rights in the EU legislatively.
It is, the, that’s just why a lot of AI services are not available in the EU. And that’s the trade off that the EU has made is to say they will, they will be technologically behind in favor of individual rights. other parts of it, like Japan and China flip the other way on that say training data is not an infringement and therefore, you know, they’re encouraging model makers and tech companies to invest heavily in their ecosystems to leap ahead and achieve technological dominance.
Some of the Chinese models like the Quinn family of models are incredibly capable. Because China’s like, yeah, IP rights are irrelevant. You can do whatever you want. for individual creators, you have to weigh the balance of, do I want my data being used to make someone else’s tool better that may infringe on my ability to earn income with.
do I want my data to be used to train the model to tell who I am, what I do, and hopefully refer people to me. It’s kind of like what we’ve been doing with SEO for 25 years is to say we gave we implicitly gave Google rights to look at our content in exchange for sending us people to buy our stuff from with and this is this is important because least according to a crystal laser, who’s a professor of law at Cleveland State University.
We won’t have an answer in settled law in the USA till about 2035. Once everything works with all the courts and all the appeals and things like that. So until then, you have to make the decision for yourself about where that line is, with the acknowledgement that those people who are more permissive in the use of their work will gain a faster advantage than those who are not.
The other thing to keep in mind is, with with certain services, you have to read the terms of service when you sign metas terms of service, you give them blanket permission to do whatever they want with your data. If you are a visual artist, and you load your images to Instagram, you have given meta permission to train models with it.
Period. End of story. If you do not like that, that is fine. We respect that, right? You cannot put your content on Instagram. If you are a musician, and you do not want Google to use that data to train a music generation model, you cannot put your work on YouTube, period, because the terms of service are what they are, they are a legally binding agreement.
And if you sign that agreement, you’re bound by it, and you cannot take it back. So from an IP rights perspective, you have to do your due diligence to figure out either where can you market that does not involve training. or it does not involve a terms of service that basically says you sign your rights over to someone else.
Michelle: Wow.
Christopher: That’s a lot to think about. I should also say, I am not a lawyer. I cannot give legal advice. Please consult a lawyer in your jurisdiction for advice specific to your situation.
Michelle: I think Ruth Carter is our lawyer go to for a lot of this stuff. Yes,
Christopher: they’re fantastic.
Michelle: Yes. Gosh, I don’t even know where to go from there.
Christopher: Well, the action plan for a communicator is number one, understand, how search is evolving. Number two, figure out where, where model makers are getting their data. three, put content there as much as possible in as many places as possible. And then for, check your results, ask your audience. Hey, how did you hear about us?
You can see in tools like Google analytics when some comes from perplexity. you can see you can, when you ask people, Hey, how did you find us? Right? If no one ever says chat, GPT or Gemini or whatever, but then you know that your audience isn’t there yet. But given that Google is shoehorning AI answers into everything, right?
There’s a good chance that even some of your Google searches are impacted. So you need to be in those places. Here’s the good part. The good part, if we are completely and totally wrong. about this. Oh no, I’ve put content all over the internet and as many places as possible. You will still get benefit from it, even if this is completely wrong.
Michelle: Yeah, I think it’s so hard to adjust for a lot of people. This will be great for people who don’t have ethics or don’t worry about having ethics. I mean, there’s a lot of those people, let’s be honest. More and more of those people. And that’s kind of what it takes sometimes, I think they think, to get ahead.
So I feel like the people that are, really experts will rise to the top out of all of the, you know, stuff that’s going on down here. But like, I, you know what, I don’t. I don’t know for sure. And then how do we, does it make, are we selling out? Like if we decide to just give it all away like that, I mean, I guess we kind of are giving it all away.
I guess you’re right. We have been for awhile.
Christopher: We have been for decades. We’ve been giving it away. because that’s the bargain. That is the whole bargain of content marketing. We create content for free so that people share it with each other. and, and recommend us and refer it. So it’s no different than we’ve been doing is it’s not an ethics thing.
It’s a, it’s a practicality thing. and in terms of, of, you know, does the best win? No, of course not. scale wins. once you have a minimum level of quality scale wins. Here’s a real simple example is. A McDonald’s Big Mac, the best highest quality burger in the world. No, no, it’s not. It is. It is an okay burger.
It’s not great. It’s okay. But it’s predictable. And it’s highly scalable. If you walk into a McDonald’s in Belgrade, you walk to McDonald’s in Manhattan, you get a Big Mac, it’s pretty much going to be the same thing. Why? Because it scales. the best has almost never won in the world. It is, it is scale and process and efficiency, which are things that AI is really, really good at.
And so good enough at scale wins over best but unable to keep up. And that is a consideration when we’re using these tools for for content creation or whatever. It is about Can we scale our efforts in, in sensible ways that meet customer needs? Cause at the end of the day, the buyer is still human for now.
Can we can we scale our efforts? Can can one person do the work of 10 or 20 or 30? The answer is yes. And the less you care about quality, the more you could scale. So if you if there’s a minimum level of quantity of quality, you can generate a tremendous amount of quality. But these tools are incredibly capable now.
at generating both quality and quantity. if you look at Gemini, Google’s Gemini Flash 002 model that just came out about a month ago, it is more capable on benchmarks than Gemini Pro 1. 5 was five months ago, which is insane to think about. that the fast cheap model is as capable as the expensive model was five months ago.
So if you are a content creator, and you’re using these tools, suddenly, you can generate stuff at a much greater scale. I’ll give you an example. I was starting to write a new book. And I was like, why this is stupid. Why should I have to write a book when I’ve already written all the book content? I just it’s just all over the place.
Michelle: Right?
Christopher: If five months ago, I gathered up all the data, my YouTube transcripts, my newsletters and stuff, I put Gemini pro and it was a arduous process. It took a few hours to generate the first draft of a manuscript. I use Gemini flash the new version and it took 30 minutes to assemble a manuscript and I’m telling it.
plagiarize from me, steal from me because it’s me. It’s my contents, only my content. I already did all the work of making, you know, a YouTube video a day for five years now and newsletters on the weekend. I already did all the work. So all you need to do is assemble it from the pieces I’m giving you. And the fact that it can do this and put together a book that sounds like me.
it because it’s using my words, as much as possible in 30 minutes is fantastic, right? Why wouldn’t I use that to scale to be able to generate one or two or three or five books? Because, again, I already did all the work. And by the way, from an IP rights perspective, now again, not a lawyer, consult a real lawyer.
If I can show lineage and provenance, I can document that this chapter, this page came from this YouTube video,
then
it is a derivative work and under, at least under most copyright law, a derivative work inherits the copyright or the original. So if I, as the human did the original work
and I can
show my evidence that the, the AI generated copy is derived from my original, I retain the copyright.
on the machine generated one as well. So if you are a public relations professional, and you’re like, Oh, we want to use this stuff, but a purely AI generated content has no copyright, that’s a danger to our business, we’ll then start with human generated work and derive it if you are working at a brand, and you have a CEO, and you can’t get that CEO to sit down and write, say, I just need you to leave me some voicemails, right, just leave me some voicemails.
And then we’ll get those transcribed, we’ll have machines infer from it and create content from it. That will be nice. It will be bespoke. It will be in your voice cause it’ll be using your words. We’ll just clean up the grammar, with machines and you retain the copyright because you’ve got the originals.
You can say, Hey, Hey, you know, this blog post by, you know, John CEO came from this voicemail message. And therefore we retain the copyright on it. Or you can hire a ghostwriter. You can hire a ghostwriter, but they’re compared to a machine. They’re slower. And if you care about scale and speed, you need to be faster rather than slower.
Michelle: I don’t like that though, Christopher. I don’t like it. Why not? Because I just feel like it’s a bunch of You know, just, I don’t know. I don’t think it’s the same.
Christopher: You did the work though. You had a human being generate the original and you’re just using the machine to clean it up.
Michelle: I don’t know. I feel like we’re just losing a lot by turning it all over.
You know, like we’re not, you know, we’re not, I don’t know. I don’t know. We’re taking the humans out and humans, you know, improve by having conversations and thinking and critical thinking and writing and reading. So, I feel like we’re really, that’s. That’s where I’m a little,
Christopher: I think if your experience has been, you’ve gotten unsatisfactory results from, from machines on content generation, a big part of it probably is because two reasons, number one, to be, to get machines to generate really good content, you need to be very good at prompting, you need to be very good at helping them understand your specific writing style.
all these tools are probability based tools, right? They’re just trying to predict the next word. If you watch what happens behind the scenes, it is literally trying to predict the next word. If it doesn’t have rules about how that word prediction occurs, then it’s going to predict the general high probability words which is means it’s going to sound pretty generic, it’s going to sound pretty absent of character.
If on the other hand, you have a machine, look at the list of 20 different characteristics of writing style, diction and pacing and so on and so on and so forth, and write out a set of extensive rules. This is how Michelle Garrett writes. And these are some writing samples, it will faithfully replicate your writing style very, very well.
Because it has rules. The way I tell people to think about machines is they’re think of them like as really overenthusiastic interns on the first day at work. They’re very smart. They have PhDs and everything. They have a PhD in in 20 different subjects, but it’s still the first day they don’t know where the restroom is.
They have no idea what your company does. And so you need to provide all that guidance to get them to to behave the way you want. You would never, never tell the intern Hey, ghost, write me a blog for my CEO. And then that, and that be the instructions, right? You’re going to get a disaster. Instead, you would say, here are 20 examples of our CEOs writing on the rare occasions where we got him drunk enough to write something down.
now from this infer and here’s the CEO’s writing style now create new content from this, that is. the way to get these tools to generate stuff that is going to be uniquely you. And you also have to have it do some level of reasoning to say, what are low probability ways of expressing something. So for example, creativity and writing and content generation comes from low probability stuff, the same thing is the same as humor.
So if I said, you know, someone was suffering from extreme gastric distress, that’s a very high probability way of saying something, it’s very very bland, very boring. If I say, Hey, it looks like your power wash your toilet with Nutella, right? That is a very low probability way of expressing that. And it’s creative because it’s low probability, you were not expecting me to say that.
As a result, if we can teach a machine, and you can how to write with low probability expressions, it’s going to sound more creative. And so the issues around using this for Not just search, but, creating stuff that is decent is, is part of prompting. It’s part of learning how to use the tool as well.
Michelle; Well, I don’t know. I might, I thought I was going to walk out of here with fewer questions, but I might have more questions.
Christopher: Exactly. and to Kate’s question in the comments, environmentally, here’s the thing. Every, not every model is the same. The big frontier models like GPT, like open a eyes. Oh, one.
That thing is a pig. That thing has uses so much compute power. It’s 30 times more computation intensive than GPT for Oh, there are so many models to choose from. Once you start getting more advanced in your use of AI, there are over a million different models that are available to use. Some of them like Mr.
All Nemo and Mr. All small, models from meta the llama models, they run on your laptop, right with no internet, no connection to anything they can run on your laptop. And so from a power consumption perspective, a resource consumption perspective, they’re very small I can run The newest version of llama on my phone, right?
That’s five Watts of power. That is a very, very small, the new Apple intelligence. That’ll be rolling out for iPhones at the end of this month, the new versions of Google Android all have onboard AI for these devices that use. tiny, tiny amounts of power. Not every task needs the biggest and fanciest model.
If you’re trying to summarize meeting call notes, right, you do not need the biggest model. And so a lot of the web interfaces like Google’s Gemini, chat, CPC, typically you have invisibly a router in front of them that will try to route your task to the model that is frankly, cheapest and fastest for them and lowest power for them because it costs them power.
Everyone and their cousin in AI right now is losing money on AI. Right? open A. I. Is burning billions of dollars a year. they are. They are not a sustainable business. And so it is in everyone’s best interest to make these things as environmentally sound as possible by using the best model for for any given tool.
Now When you look at Google’s AI overviews, for example, it’s pretty clear they’re using Gemini flash, right? The lower power, lower cost model because it has to scale so fast. The same is true for Gemini in Google Docs, Google Sheets, etc. They’re using the low power model. And it shows because it’s not as creative.
It’s not as as as thoughtful as the more powerful model. The same is true for all these tools when you use Bing. Yeah, it’s using GPT for it’s using kind of a dumb version of GPT for because it’s lower power.
Michelle: Goodness. Well, let’s see what Kate is saying here. The argument. Okay. Let’s just put this up. The argument after that is a human intern or ghost writer will take longer, might do a worse job, but keeps a human in a job. So you’re using them because you believe that that’s right. Despite it being more efficient from, from a business perspective to use AI.
But that’s fine. That’s fine. But it’s a trade off. See that that’s kinda, I think that’s exactly the essence of
Christopher: exactly. So, and here’s a great example of this. It takes a human being 10 hours, 10 hours to harvest a bushel of corn by hand, right? It takes the same human being in 10 hours can harvest 23, 000 bushels of corn driving a john deere x 9 1100.
Right? That farm does not employ 1000 people anymore. That farm employs that same one person who just drives the machine and harvest 23, 000 bushels of corn. Do we have a moral obligation to go back to farming with 1000 people, some of which many of which, you know, in the old old days, were worked with under absolutely abhorrent conditions, including slavery, right?
Or do we go with the new version of mechanization and automation where it’s one dude, you know, with his with listening to his podcast as this GPS guided, massive house size machine just goes up and down the field? Which do you choose? Right? Do you have I like that analogy with the case bringing because it is a situation where like, yeah, that automation has huge benefits, less human suffering.
less, fewer mistakes, right? Because it’s all automated, but the trade off is less employment far less far that we went from about 70 percent of the population in the USA, working agriculture to less than 1%. And yet we feed more people than we ever have before. Now is knowledge work the same? Maybe is can be made more efficient?
Yes. Can we then have those same individuals were be able to do 10 or 15 or 20 or 100 x more impactful work? Maybe it’s early days, in the same way that no one was really sure how the steam engine would transform agriculture, or the internal combustion engine. We don’t know how AI models will change things, except that we know that in general, fewer people will be needed on for an individual job and more, and people who are in that job will have more, we’ll have very different looking tasks.
They will be piloting the machines in the same way. The farmer is driving the John Deere, you know, mega combine. And how does that work for, for employment? We don’t know.
Michelle: Wow. Gosh, again, I was hoping to walk away and feel better about this, and I don’t know if I do. Oh my goodness. well, is there anything else?
I know we’re kind of winding up here, and I really appreciate Kate’s questions. If anyone else has a question, please feel free to post that. And I will be sharing this afterward, of course. But what else, Christopher, do you think we should be focused on? Is there anything else?
Christopher: there are so many tasks in marketing and communications that really should be automated.
So I’ll give you an example. One, I used to work at a PR agency and we had these account coordinators copy and pasting Google results into a spreadsheet. That’s not even an AI task. That’s a basic automation that that task should not be done by humans. Another one making call transcripts and meeting notes and stuff that should not be done by humans that should all that should be handed off to machines so that the humans can sit there and think through.
Okay. So we have. our conversation with our client, what should we do about it? What kind of human thinking and strategy? Could we apply to that information? Right. That is true for all of this stuff in PR and communications is people have gotten so used to making the stuff, right? Okay. I’m going to make this press.
We’re going to make this article and stuff like that. Instead of is this, what should we, should it be making, or is, you know, can we let the making part be done by machines so that we can get to the thinking part,
Michelle: but I’m going to make a point here. I feel like the process helps you. It does help you think and get to where you need to be.
I do agree with you that too much stuff is made without asking questions. Why are we making it? Who are making it for? I absolutely, a hundred percent. But I think part of even going through my notes after a meeting or, you know, sending a summary to a client helps me organize my thoughts to help me do a better job for the client.
So I feel like, again, I wouldn’t want to turn that over, but. Okay. That’s, that’s, that’s my point, I guess.
Christopher: And that’s how we work, right? That’s how people in our generation work. People who are, you know, in generations after us work differently. Ask anyone under the age of 30, who has not been in the military, how to use a paper map.
Right. Ask people, you know, anyone under the age of 40, if they can remember the, their five, their five closest friends, phone numbers, we don’t remember that stuff anymore. We don’t know it anymore. Do we need to, maybe not. and so that is the question now. to to wrap up on the AI based search stuff.
Here’s the thing that we just have to keep in mind. We have to be everywhere. You have to be everywhere. You have to generate content everywhere that the machines are looking. And you may need to use machines to do that, to scale that much, because your client probably is gonna say, yeah, I’m gonna add a zero to your PR budget so that you can, you know, 10X our program.
They may say, yeah, we’re okay with the retainer that we have, but we need you to do 10 times the amount of work.
Michelle: Yeah, I don’t know, we’re going to have to keep talking about this.
Christopher: Mm hmm. That’s correct.
And I’m going to make sure I get all my, my episodes transcribed, they’re on YouTube, but I got to put a page up on my website now with all that.
Christopher: Mm hmm. Yes, you should. And you should be using generative AI to clean up those transcripts to fix the grammar.
Michelle: Well, okay. I think, I think, I think I can go that far. Is there anything else? Tell us how to keep up with you. What’s the best way for people to keep up with you? Your newsletter? Where are you going to
Christopher: TrustInsights. ai. That’s my company. and you can find, you’ll, you’ll be able to find your way to all the, the socials and other stuff in there, but go to TrustInsights.
ai. That’s where you can find me. And if you want to find my personal stuff, you can find that at ChristopherSPenn. com.
Michelle: All right. Well, I really, I can’t thank you enough for being here. You’re one of the smartest people I know, and again, I’ll be thinking about this, and maybe I’ll even write about it, and you won’t mind if I take your quotes and put them in my blog post.
Christopher: So you certainly can try using AI on it.
Michelle: All right. Thank you. Thanks, everybody. Bye.
