AI Writer and AI Design Tools for Marketers

- What's happening is, as you're, say writing a blog post,
you can, like Jasper is an example,
as you're writing the post,
you can hit generate image and it can actually generate
an image relevant to that paragraph.
(funky electric music)
- Today I'm very excited to be joined by Paul Roetzer.
If you don't know who Paul is, he's a marketing AI expert.
He's author of "Marketing: Artificial Intelligence."
He founded the Marketing AI Institute
and the Marketing AI Conference.
He's also co-host of The Marketing AI Show,
and his course is called Piloting AI for Marketers.
I sense a theme here, Paul.
Welcome back to the show.
- Staying on brand, aren't we, Michael?
- How you doing today?
- I'm doing great.
It's so good to be back with you.
It's been a while.
- Well, yeah, it's been a couple years.
I think it was about two years ago and a lot has changed
and today Paul and I are gonna explore AI tools
that are gonna help marketers better create images
and write better and a whole lot more.
Paul, I think one of the big questions
that a lot of people have on their mind
right now is
why they should embrace AI?
So may many of them see AI as a threat.
And I would love you to instead make the business case
as to why AI is going to help marketers
and why really they should embrace it.
- The two most obvious things to me are
the efficiency with which you produce creative.
So if you're a writer, if you're a designer,
you're a videographer, whatever it may be,
whatever your medium is,
you can produce the content more efficiently.
And then unlocking creative potential.
You can be more creative as a professional.
So it can then, if you're a good writer,
it can make you a great writer.
If you're an average writer, it can make you a good writer.
If you have no design capabilities, like me,
you can now create your own images.
Like, it just opens up new possibilities
of efficiency and creativity.
- Yeah, I was chatting with a friend who is a copywriter
that back in the day I used to be a professional copywriter
and what I said to him is,
imagine if you could 10X your output.
You could lower your fees, right?
And you could gather more clients
and you could ultimately make more money in volume.
The other side of the equation is I do think
that these AI tools that we're gonna be talking about
are going to improve communications
and visuals everywhere
and it's gonna kinda level up everything.
And I use the analogy, did Photoshop kill art?
Did Illustrator kill art?
No, it didn't.
Actually, it created a whole new frontier of artists, right?
who were, instead of traditional painters,
or traditional photographers,
they were able to do things they could never possibly
have dreamed of until these tools were invented.
I'm just curious what your thoughts are on that.
- No, I agree 100%.
The way I've explained it is
AI is not gonna replace anybody,
but the professionals who adopt AI
will replace the professionals who don't.
So, you know, it's not the AI's fault
that some people skillsets will be obsoleted,
or their roles or career paths.
It's the ones who just refuse to embrace that.
Yes, technology's evolved.
There's new toolsets to now do things,
whether it's like you explained with Photoshop,
they're just new, smarter tools.
These are just extremely powerful tools
that put these capabilities into the hands
of the average marketer or creative professional.
And that's the difference.
Like, I still can't use Photoshop.
It didn't make me a designer.
A DALL-E does, like I can now.
So I think,
the big change in 2022 was the advancements
in the generative AI that made AI accessible
to the mass market of business professionals
and marketers and creative professionals.
Anybody can now use these tools.
- Well, and a couple other business advantages
are they never sleep.
You don't have to wait for some person to have availability
in their schedule for you to hire them, right?
You can literally go for very, very low cost, generate copy,
generate images that could be used literally moments later
in all sorts of different capacities and I kinda feel.
We're recording this, literally,
like on one of the last days of 2022.
And I would love you to talk a little bit
about why, all of a sudden, right now,
things seem to be exploding.
And talk to me a little bit about this concept
that this is just the training ground.
- So generative AI is the term
that's caught on toward the end of 2022.
It really started with DALL-E 2
when it came out in the summer.
So it was introduced in April,
but it was readily available
to the wait list people in June and July of '22.
That made it so that anyone that had access
could experience AI in a new way and create something.
That magic moment, eight seconds after you gave it a prompt,
you had six images that were more than you could imagine.
That was a turning point.
Then the real pivotal moment came
when ChatGPT was introduced on November 30th.
Within four days, more than a million people
had accessed and tried ChatGPT.
So now language generation was accessible to everyone.
And those two things really shift
and now there's a lot going on in other companies
we'll talk about better players in this,
but those really shifted the accessibility of AI
and the understanding of its power and its potential.
Now where it goes is the interesting part
because ChatGPT and DALL-E, in some ways,
are really just training grounds for what comes next.
ChatGPT, in particular, they come from OpenAI,
which is an AI research lab,
but ChatGPT is kind of a middle version before the next.
So GPT 4, which,
may be out when people are listening to this later in 2023,
we already have GPT 4,
but what's happening right now is all this user feedback,
what's called reinforcement learning with human feedback.
As you and I are writing things in ChatGPT
and be like, oh, that was awesome,"
I'm clicking the thumbs up or I'm continuing on,
it's learning that it's doing things well or not,
and they're able to adapt it
and then infuse that learning into the next version,
and that's the one that's gonna change everything.
This is just experimentation.
- What year did you found your AI business?
- So 2016,
it started as a DBA,
a blog basically under my agency, PR 20/20.
And then in 2019, January, we split it off as its own LLC.
- So given the fact that you've been tracking
this space since 2016,
what do you think the importance of the moment
that we're in right now is for the business world
and the world in general?
- Yeah, so I started researching AI in 2011,
trying to understand its application
to marketing into my agency
and we started writing about it in 2015,
launched the site, 2016.
The way I explained it is, like, back in 2011-2012,
once or twice a year something would happen in AI,
not in marketing 'cause it wasn't really being
talked about in marketing, sales, service, business yet.
But in the marketing or in the AI realm,
or the research realms,
once or twice a year there
was like a breakthrough worth talking about.
Then like 2016, '17, '18,
it was like once or twice a quarter
something major would happen.
Like AlphaGo, when DeepMind built AlphaGo,
that was a major moment in AI in 2016.
Then in the last couple years,
it was like once or twice a quarter,
once, twice a month something started happening.
In the last half of 2022,
it was like two to three times a week
something was happening.
Some generative ad company was getting
a hundred million in funding,
somebody was releasing a new model,
somebody was releasing some major breakthrough
in reasoning or whatever it is.
And if you start to connect the dots,
what you realize is now there's a bunch
of really practical tools for marketers.
Like we can go get these tools right now
and play around with them,
but it's what's happening behind the scenes
where most people aren't watching in the research labs
and like what's kinda coming next,
that you can look out in 6 to 12 months
and start projecting the real disruption that is near.
And that to me is why this moment is so unique
is it's accessible to everyone,
so we can actually experience and understand it
and we are at a profound shift
in what it's going to be capable of doing.
- Yeah, and marketers pay attention
because I believe, as Paul mentioned earlier,
that it's the marketers that embrace these tools
that are gonna have a unique competitive edge
and the ones that fight against it will not.
So let's start with design.
First of all, before we get into the actual tools
and you hinted at DALL-E,
we'll get to DALL-E in a little bit,
let's talk about some of the practical use cases
that maybe people listening right now don't realize
you can accomplish with these simple AI design tools.
- The simplest thing for me is think
about anywhere where you use images today.
Could be collateral, could be your blog posts,
social media shares, ads, wherever you're using images.
And maybe you're getting stock photography,
maybe a license to some library,
maybe you're using original photos, whatever it is.
Moving forward, any text you create,
AI can create an image to go with that text
and you can eventually train those images
on your brand guidelines.
Like, we want this certain style,
this certain theme of our images.
But what's happening is,
as you're, say, writing a blog post,
you can, like Jasper is an example,
as you're writing the post, you can hit generate image
and it can actually generate an image
relevant to that paragraph
that you want the image related to,
or I can output-
- Wow.
- My 20 social shares and say, create images for all these.
So again, anywhere where images
are living in your marketing today,
you're going to be able
to generate original images to fit prompts,
and that could be the existing text
or some other text you give it.
- Yeah, so obviously for this show,
Social Media Marketing Podcast,
the obvious stuff is images that you might want
to use on the various social platforms.
Then, of course, you mentioned blog posts.
I could see header images, right?
Like when you have a big image at the top of the blog post
and maybe you just Photoshop some words over the top of it.
Stock images obviously could be huge, right?
And I start to think about those of us
that are presenters and like to use images.
And we should state, these are photorealistic images, right?
- It's whatever style you want.
You can say 3D render, photorealistic, chalk drawing,
you tell it what the output is, but yeah, 100%.
I did my Marketing AI Conference keynote in August
and I built the entire first half of the deck with DALL-E 2
and I didn't tell anybody I had done it yet.
And then when I got to the middle part,
it was like the unveiling,
like, hey, here's the practical use.
Every image you saw, and I showed a collage,
was generated with one single prompt.
I didn't even edit these things.
So I built all these images in like 30 seconds.
- What about logo design and vector images, right?
'cause most of this stuff is app images
that this stuff is creating, right?
- Yeah, so the vector images is an interesting realm.
I'm not sure of any of 'em
that are doing that at the moment.
Again, by the time someone might listen to this episode,
they're probably going to be.
That's like a next iteration where you can download
a high-res version that you can actually edit.
But images, like logos,
there was a company that was doing that before DALL-E 2.
So my guess is you're gonna see major innovation there.
So anything you need.
Like right now,
they struggle with with words, with characters.
So if you wanted to do a logo of your company
and you wanted the company name in it,
it's not gonna do that very well.
It's gonna jump up the letters and things like that.
- But maybe get you an icon
that you could use next to your name,
right? - 100%, yep, yep.
- So one of the things that a lot of people
are probably thinking about is,
okay, this is really fascinating.
What about like copyright, rights management, trademark?
And I know that this might be a it depends,
but let's talk a little bit about like the use cases,
at least with some of the tools that you've used
and we're gonna get to all the tools in just a second,
but generally speaking, what can we do with these images?
Since we are marketers,
we might wanna use some of these
and paid promotions and stuff like that.
Are there any slippery slopes we need to be aware of?
- So in the near term,
the simple answer is the images are original images
and assuming the vendor you're using,
again, OpenAI, we'll stay on DALL-E 2,
when it first came out,
they did not give you commercial rights to the images.
That quickly changed.
By like month three, they changed where you now have
commercial rights to use the images.
So you're clearing, clean.
It is not a scraped image, it's not a stock photo.
We could get into how this stuff works,
but uses something called diffusion where it actually learns
from a bunch of images on the internet
and then it generates pixel-level-up original images.
So it's yours.
And if you and I gave the same prompt to DALL-E 2,
we're gonna get a different image.
So it may look similar, but it's not gonna be the same.
Language is the same.
It's not scraping content.
If I give it prompts,
they write me a blog post about 10 trends
in social media for 2023, it's going to write it.
If you run that through a plagiarism checker,
it's going to pass it 99.9% of the time.
So the copyright on that, in theory,
you could copyright it.
So as a individual user, it's pretty straightforward.
The complex part becomes,
we don't have precedent yet legally around the training data
of where this original images and language comes from,
because it learns from the internet.
So I could write a post right now
that says 10 trends in social media for 2023
in the style of Michael Stelzner
and it could write it like you
'cause it could go and learn how you write.
Now, could you come at me and be like,
dude, you just, like, stole my voice.
That's where the issues are gonna come in,
because the image generation is trained
on images of the internet,
which is someone's original work
and language, in a similar way,
but the language actually works by predicting the next word.
It's not actually writing anything it's seen before, per se.
It's not, again, scraping or summarizing,
but I don't know how far we wanna get
into how this stuff actually works.
- No, no, I mean, that's totally cool.
And I think, you know,
we had met when we were prepping for this show, we said,
for sure, look at the terms of service
of whatever tools you're using.
But you also mentioned that you can upload
some of your own stuff to train it.
Talk a little bit about how that might result in something
even more secure, if you will, towards your,
unlikely to create any kinda problems.
- Yeah, so if you're a media company,
a publishing company, a larger brand,
and you have a team of writers, like one way to do it,
like writer is one of the vendors that does this,
you can train it on your style guidelines,
your brand identity and style guidelines, your tone,
your voice, the specific ways you say certain phrases.
You could do it across multiple languages where you can,
you know, actually say, like, set guidelines about it.
So you can set some parameters, and then, in theory,
not just with writer but in others, you'll be able to say,
hey listen, here's my last hundred blog posts I've written,
I would like you to learn how to write in my style.
So it's not now just some general AI writing tool.
It's a tool trained in the style
of specific writer or team or brand voice.
So it's not available right now accessible through,
you know, the affordable tools we can all go get.
It's going to be, like if you're Chris Penn,
like, you know, mutual friend Chris Penn,
Chris knows how to train that stuff.
Like right now you need some coding ability
to go in and actually build a model yourself,
but that's gonna all be kinda baked into these platforms
and you're gonna get our AI writing tool
and you say, train it in this voice, and away you go.
- Let's talk about some of these actual design tools.
Let's start with DALL-E, spelled D-A-L-L hyphen E.
What do we need to know about that?
And then let's talk about some of the others as well
'cause you mentioned DALL-E quite a few times.
- Yeah.
So DALL-E was the first one to really hit the market.
Google has one called Image Gen, but it's not available yet.
DALL-E is I think $15 a month,
15 dollars a month when I was using it.
It gives you certain credits to how many,
images you can create
and I think you get the first 15 free.
I haven't looked at their pricing in a while,
but I think I paid $15 a month for DALL-E.
So you can have an AI image generation tool for $15 a month.
A Midjourney, go ahead.
- Well, let's mention briefly the prompts, right?
Like explain how that works and how the prompts,
for people might not understand that.
- Yep, so the key to generative AI,
is you give a prompt to the AI and it generates something.
It's a very literal term.
So prompting is just giving it a set of instructions
of what do you want it to do?
Write me a blog post,
create an image in 3D render style of this.
It's telling it what to do.
- You can be super descriptive, right?
You can say with sky and palm trees.
- Yeah, you can go as deep as you want
or you can start off general and just like give it
a very simple prompt and see what it does.
So right now there's a term called prompt engineering
and that is literally engineering the prompt.
And so in design, this is where I feel like the artists,
the designers who use AI have an advantage,
because they're able to explain what it is they want
visually better than I, a non-designer, can.
So they know styles, they know shapes,
they know the things that they want
to prompt the machine to generate.
So they're likely going to get richer outputs
from the machine than a non-designer or artist would.
Now when you go into language.
I'm a writer by trade, you're a writer by trade,
we probably have an ability to better prompt the machine
with what it is we want the output of the writing to be.
And so prompting is literally just saying,
this is what I want you to do.
Create 10 social media shares,
with 150 characters, or less.
That's what a prompt is.
- Perfect, and we'll get into prompt tips
and stuff a little bit later.
So DALL-E, well Midjourney, let's talk about that.
- Midjourney is another one.
People have done some incredible stuff with Midjourney.
I haven't personally used Midjourney,
but, I mean, it's mind-bending.
I've seen in like illustrations, video game designs,
like just crazy stuff that I've seen done there.
- What's the difference between DALL-E
and Midjourney as far as,
do they output tiny little images or high-res images
or can you choose or how does that work exactly?
- It depends, like the bigger the image,
the more it's probably gonna cost
because the more compute power
to do it and things like that.
Midjourney, seems like it has,
it's meant to design more expanded images and generations.
Stability AI is another major player.
They just raised 101 million.
RunwayML is actually one of my favorites, to your question.
Like, so it's just, I think it's runwayml.com is that one,
but they have one called Infinite Image where I can create,
just like I would with DALL-E, I go in and give it a prompt,
like a hilltop with stars above,
and it'll generate an image and then I can
actually drag a box and I can continue that image,
like, now, put an observatory over
and then I can, like, and it can just keep going.
So you can just.
- Does it layer it on top of the existing image?
- Yeah, seamlessly.
It's friction, it's beautiful.
Like it's insane how it does it,
like the blades of grass continue
on from the previous image.
- So this would be really powerful
for a true painter or artist who wants to.
- Oh, yeah.
- Who's familiar with maybe using Photoshop,
and I wonder if it can output these layers independently.
Do you know if that's.
- I don't know if it can, but I mean it's.
- You could as a Photoshop file
or something like that, right?
Move them around and stuff like that.
And I pay $28 a month for runway
and they have images, videos.
They're playing in texts now.
So you can generate all this stuff.
They have a whole, it's called their AI Magic Tool suite,
but again, $28-a-month like capabilities
like I would never have
because I have no design capabilities.
- So if somebody was to.
Okay, so we talked about DALL-E,
Midjourney, Stability and Runway.
Obviously, they've been trained
with different sets of art, right?
So they have different styles.
Is that why they're different?
- In theory, it's the models and how they're built.
So like the language models,
OpenAI is the primary provider of language models today.
So a lot of the AI writing tools you would buy,
like a Jasper, Writer, HyperWrite,
they're built on top of OpenAI's API.
So they're using that language model
as the core of what they do.
So similar if you're using like a Jasper Art,
I believe they're actually using the OpenAI API for DALL-E,
like they're actually just piping in DALL-E's capabilities.
So they're the companies that are
just using kinda the platforms to build on,
and then there are the companies like Runway,
that are building their own models.
And so that's the difference,
is some of them are actually building their own.
Like Cohere in the language space,
is building its own language model.
So as the end user,
like I don't think most marketers are going
to get into the weeds of like how this works.
They're just gonna go get a Jasper
or a writer or whatever it is.
And it's like, oh, okay, cool,
it's got AI art and AI language,
and they're never gonna care the model behind it.
But the savvy designer who's like,
let's say it's a design studio,
they're gonna go out and they're gonna play
with each of the models and they may use Midjourney
for one client and they may use DALL-E for another.
And that, to me, is the opportunity is,
like take your domain expertise,
go figure out the way to infuse these tools
and maybe it's multiple tools into your business.
- Well, and I could see a day where Canva integrates
with hundreds of these services and then they have
an infinite library of images.
- Tome, T-O-M-E, I just.
- I could see them using this right?
- Oh yeah, they have to.
But Tome is one I just came across that does slide decks.
So you just give it a prompt, like build me a deck
for a startup that's raising a million dollar seed round
and it'll actually start building it
with language and images build into it.
- How do you spell that again?
- T-O-M-E.
Let me check and see what the URL is.
It's on my list to test this week.
- Well, and while you're doing that,
we should probably briefly talk about Descript as well,
if you wanna go ahead and explain.
- Oh, Descript is awesome.
- So Descript is the one we use for audio and video.
Again, I think that license is under $30 a month
or maybe it's like $16 per user, per month.
Super affordable for what it does.
So they do things like clean up your audio.
So if we take our audio from the recording,
whether it's Zoom or streaming or whatever it is,
you can upload the audio and it'll clean it up.
It'll remove noise, remove, kind of that stuff.
It'll do transcriptions, it'll do it.
It'll do speaker recognition so it knows who's speaking.
You can train it on spellings and things like that.
It has an a filler word remover.
So it'll actually just click this
and it'll go through the entire video
and just remove all the filler words.
You can do synthetic voice.
So if like, let's say I forgot
to say something on this podcast.
I'll say, Michael, I really should have said
this in that podcast,
you could actually put this into Descript,
type in what I wanted to say
and it'll synthesize my voice, saying that thing.
- And it'll sound just like you.
- It'll sound just like me.
Yeah, it's like wild stuff.
But Descript's crazy.
Like if you do videos, podcast webinars,
you just have to have it.
- And audio too, right?
Obviously you could just use it
for your audio podcast as well.
Did you get the name of that spelling of that name?
- Yeah, so it looks like it's, make sure this is,
this might not be the right one.
Tome.
- Okay, well, while you're looking for that, so folks,
what I think is really exciting about this,
is these tools are gonna make things possible
that you might not have been able to do before.
There are plenty of marketers listening right now
who don't have access to designers,
don't have access to writers, it's outside their budget.
You're gonna be able to use these tools to do stuff
that you could never do before and you might become,
if you will, an AI artist, right?
Which I think is kinda fascinating
because there will eventually be demand
because there will be people who figure out
how to use AI art as source material
and then layer in stuff on top of it on Photoshop
to create things that frankly were impossible
to create before.
Did you figure out?
- Yeah, I got it.
So it's Tome, tome.app.
- Perfect.
Let's talk about the writing stuff.
We've hinted at it a lot.
First of all,
let's talk about some marketing use cases
where these AI writing tools
can really help take things to a higher level.
- Yeah, I mean, again, kinda like the images,
anywhere you generate images is a possible use case.
Anywhere you generate language is a possible use case.
So think about yourself or your team.
All the places where you write,
could be emails, blog posts, social media shares,
adcopy, PowerPoints, whatever it is,
anywhere where you're telling a story or communicating,
AI writing tools can be used to enhance that and not.
I never talk about full automation of the copywriting.
You want the human in the loop and we can dig into that,
but gets you pretty far along when it comes to that.
Most of the time it excels today at short form
because it generally loses its ability to be original
and not repeat itself once it gets to,
say, 700 words or more.
So you can't just put in,
write me a novel about X and it just writes 50,000 words.
That's not what we're at today with this stuff.
- Let me give you some examples
of the ways that I've used it.
I have a bunch of people including yourself
who are speaking at Social Media Marketing World
and they submit headlines for their sessions.
I go into ChatGPT and I ask it
to create 10 variations of it.
- Nice.
- And then what I do is I use my own copywriting skills
to pick the best variation and modify it slightly.
So instantly, I get them.
We're creating an email.
So what I'll do is I will paste in the email
and I will ask it to create 10 subject lines.
Then I will ask it to create 10 subject lines with emojis.
Then I'll ask it to create a shorter,
more concise version of the email,
which is a reminder email.
And I'll say, it's a reminder email, a follow up email,
so it'll modify it slightly, right?
I have a friend who I spoke to last night
who created a 70-page ebook that he's selling on Amazon.
And what he did was he used the tool to, first of all,
come up with the chapter headers for 10 chapters.
And then he used the tool,
and then he used it to come up with subsections.
Then he used the tools to write the various sections
of the subsections.
And then what he was able to do was go in
and edit the entire thing and he had an ebook done
in a couple of hours, a 70-page ebook, right?
So you start thinking about these use cases, right?
And just from a social marketing perspective,
so many, it can do really good, tight, little prompts.
And even eventually you'll learn
that it kind of uses certain words over and over again
and that's where you've gotta get a little creative
and just use it as a source of inspiration,
but, outlines, ebooks, some.
Oh, and it can also summarize information really well,
which I think is worth talking about.
So I put a 45-page transcript into one of these tools
and I said, write me a summary, and it wrote me a summary.
So, it was okay, it was pretty good.
And you can also have it summarized
by omitting certain things.
You can say, write it in first person,
write it in third person.
So like you said,
almost anything you can imagine this thing can do.
And generally speaking,
I don't like to use it as out of the gate.
I like to use it as inspiration
and then we go back and modify it,
but man, the speed at which it operates,
we're talking like seconds, right?
It's pretty.
- Oh yeah, yeah.
And the summarization is a really interesting one
and I think you're calling out a lot
of very obvious use cases, again,
like the things you already do,
but then you layered in things you wouldn't do.
Like you probably wouldn't take the headlines of 50 talks
and come up with 10 variations of each.
So you're actually enhancing your creativity
because you understand what it's capable of doing.
And that to me is the beauty of it, the summarization.
Another one I think you and I touched on when we were
like prepping a little bit is summarization of videos.
So like put a YouTube link in
and it'll summarize the video for you
and the way it does that.
Oh yeah.
- Wow.
- Yeah, so you take the YouTube API,
you get the transcript 'cause YouTube
automatically transcribes the video.
So you take the transcription from the YouTube API,
you pipe it into the OpenAI API through GPT-3 or whatever,
use GPT or ChatGPT, whatever it is,
and summarize this link
and then it'll actually summarize the transcription.
Now it has some limitations today.
Like some I've seen are like,
anything beyond nine minutes,
it won't do well 'cause it's just too much text.
- So wait, just to be clear,
can you paste in the YouTube link into ChatGPT
and it'll do it for you?
- ChatGPT, I haven't tried it.
It probably will.
No, because it's not connected to the internet
so it wouldn't, but OpenAI's Playground probably could,
which is a backend tool.
But like I saw a guy last night
who just like created it over weekend.
Like he was just playing around and he just created it,
a YouTube summarizer that does 10 bullet points
of every video.
You just grab the link, drop it in there, hit generate,
and within 10 seconds it had a 400-word summary
of the nine-minute video.
- Yeah, and we should probably talk about the fact
that the tools that we're about to talk about
at ChatGPT and really the big one that everybody's
been talking about, their knowledge base,
I believe is right up until 2021.
Does that sound about right?
- Yeah, the training data for ChatGPT was end of 2021.
- So if you're a writer like Paul and I are
and you've got a bunch of content
out there that you've written,
you can essentially say, write about,
write about in my voice, right?
in the voice of Michael Stelzner or Paul Roetzer,
and it knows things.
So for example, I gave it an example.
I said, write me an email about a sale ending in 24 hours
for Social Media Marketing World.
First of all, it knew what Social Media Marketing World was.
That's the first thing.
Secondly, it knew what the value proposition
of Social Media Marketing World was.
I didn't have to give it anything,
but it also knew the proper model
to write a sale-ending-soon email, right?
And that's where the magic starts
coming together with these tools.
So you've talked about three tools throughout today,
Jasper, writer.com and ChatGPT.
Let's talk a little bit about each one of those
and kinda what makes each of them different.
- So I mean Jasper was the big name in the, toward,
you know, the end of 2022 'cause they raised
125 million at a 1.5 billion valuation.
- Wow.
- Yeah, so that kinda was like a bit of an inflection point
I think in the language generation space
because they've been pushing hard.
They've had a lot of growth,
but now they're,
I mean, their growth is astronomical,
how quickly they've grown their user base.
- And who are they for and what do they claim to be like?
They're for marketers aren't they?
- Yeah, it's a right, yeah.
So I think their,
I mean their pricing model is probably shifting,
but I wanna say I have their team license at $99 a user,
which is like all the features I need
and it's got everything baked in.
- A month?
- $99 a month, yeah.
- Okay.
- I think they have like a 20-some dollar
a month option so you can get in quick.
Now that one has a bunch of templates.
So the way a lot of these tools emerged,
like HyperWrite and Jasper and Writer,
they train a bunch of templates like blog post template,
adcopy template, Facebook Ad copy template.
And it's just like, there's like dozens of these things.
I mean 80, when we wrote our book,
"The Marketing Artificial Intelligence" book,
they had like 70 templates already.
And so I'm sure there's just like a hundred of them.
So you go in and you say, okay,
I'm writing an Amazon ad or I'm writing a Facebook Ad
or I'm doing social shares for Twitter
and you just click on the thing
and then you put in, like, I want on this link,
I want this description, this voice, whatever,
so those are designed to be writing tools
where you go in and you generate copy
within these templates.
ChatGPT,
I dunno if forced is the right word,
but it changed the dynamic of the market
because what happens with ChatGPT,
if you've used it is it's stupid simple.
I just go and I put a prompt.
I don't have to click around to anything.
I want a blog post.
- It's a flashing cursor really, right?
- Yeah, it's just, like, I want a Twitter share,
I want, just tell what you want.
I don't have to look around for templates.
I don't have to go through all this searching.
So within three weeks, all the major players
in the AI writing tool space,
the platforms introduced a chat function
within their platform.
So now in Jasper or in Writer,
or in HyperWrite, or any of the others,
I can just go in and have a ChatGPT style experience,
rather than clicking around to my templates.
Where Writer differs is, it was mainly initially built,
my perception is more of like a Grammarly,
where you're able to go in and actually edit, but for teams.
So I can go in and I can train, it's like, hey,
I have 30 writers, I have a hundred writers, whatever it is,
and I can actually train it to edit within the style
and voice and guidelines of my company
and you can have uniform editing capabilities
across the team.
And then they built in the writing capabilities
on top of it, and they actually released,
I wanna say it was early December,
they made the writing capabilities
like GPT-style writing capabilities
baked into their lower price packages.
Originally it was enterprise only and they made the move.
So it's, theirs is kinda like,
almost like Grammarly meets Jasper kinda idea,
I guess, is the way to think about it.
- So ChatGPT as of this recording is free.
There doesn't seem to be too many limits on it
and it doesn't,
unlike these other tools which seem to limit the amount
of input that you can put into it,
and they limit the amount of output
and they're typically based
on how much you generate from it, right?
But ChatGPT is part of the bigger technological company,
if you will, or entity that's powering
all these tools to begin with, right?
So do you recommend people start with ChatGPT right now
because it's completely free and it's the same tech
that's fueling all these other tools?
What's your thoughts on that?
- Yeah, so for experimentation,
like if you just wanna see what this stuff does
and start playing around with it,
ChatGPT is the starting point 'cause, to your point,
it is free and it is the most advanced thing available.
You can also go right to the source.
So OpenAI has a what's called OpenAI Playground.
It's a little bit more,
you can customize it a little bit.
You can adjust what's called the temperature,
which is how creative the output is.
So again, how, in a very simple way,
I'm not gonna overcomplicate this,
how a language generation tool works is it tries to predict
the next word or sentence based on the prompt it's given.
So if I wanna say go and sit down in that chair,
the probability that chair is the next word
is probably the highest probability.
Well, if I adjust the temperature,
it makes it more creative.
So it may output go sit in that on the floor,
or on a llama, or like whatever it is.
So the engineers,
the people kind of in behind the scenes
playing with these models can adjust these settings
and there's like three basic settings
that determine the probability of the output
being what you think it would be,
versus like extremely creative.
And so in Jasper and those,
like the user design tools, you don't have those.
Like I can't go into Jasper and adjust the temperature
and play around and see what it does.
But if I wanna see it,
if you're someone who actually wants
to understand how it works,
then go get OpenAI's Playground or Cohere is another one
and you can actually learn how the language models
are doing what they do.
- What does GPT stand for?
- Generative?
That's a great question.
Generative something transformer.
- My understanding is there's two
different algorithms at play here, right?
Like there's an algorithm that generates options
and there's another one that picks the best option.
Is that correct?
Two different AIs that are kind of like working together
to be able to output something that's really great
or is that just not.
- It depends on the setting,
generative pre-trained transformer?
So transformers were invented by Google in 2017.
It came out in this research paper
and that became the driver of the major advancements
within language generation.
So if you really wanna understand this,
go back and read the original Google paper
'cause transformers is what enabled OpenAI
and Stability AI to do what they're doing.
They're the thing that powered
this whole generative AI phase that we're seeing.
- What's cool about this whole chat interface
is for those that have been listening
to the show for a very long time,
we've had like the founder of Manychat
on the show I believe, if I'm not mistaken.
And we've had chat,
the traditional chat in the social platforms
was very much based on tree structures, right?
Like if someone types in this word,
it was, if this, then that,
and you had to build all this stuff, right?
And there was no natural language processing tools
that were available, right?
which is I think part of the promise
of this chat functionality is now, all of a sudden,
I could see having a customer service entity input
all the answers to all the most commonly asked questions,
right, inside my company.
And I could see this chat concept
being real live customer service.
And then when I start adding voice on top of it,
I could see call centers that are powered
by this kinda stuff as well.
What's your thoughts on that?
- No doubt and, I mean, you can do it now.
So like, you could, for example,
give it training data of all the best responses.
So like you could go through all the call logs
and say, these were great responses,
and you could, in theory, train a system on that.
What I've seen people doing now with customer support
is like, okay, I'm gonna respond to this upset customer
and rather than me just sending my email,
I'm gonna go into ChatGPT first and say, okay,
how should I address an unhappy customer
who wants their money refunded and put in a prompt and like,
oh, that's actually a pretty good email,
let me use that one,
or like do a variation of it, so you can actually start
to use it to inform the decisions that humans are making.
But in the not too distant future,
I could definitely envision conversational agents
or call centers where the predominant
or the first response is actually coming from these agents.
And again, it's not new in conversation,
there's incredible companies,
like LivePerson used to be a client of ours
way back in my agency days.
Drift, like there's major players
in the conversational agent space
and I think what they're going to be capable
of building now is going to significantly improve
on what they were building before 2019-2020.
- Now ChatGPT or OpenAI,
my understanding is Microsoft paid
over a billion dollars for exclusive rights
to be able to license this technology, right?
So this is not gonna be a free forever
kinda thing obviously, right?
- No, and there's no official number yet
of what ChatGPT is costing OpenAI,
but I've seen 3 million-plus per day,
because basically each prompt you give costs them money
to power the compute to enable that response.
And it's probably somewhere between a penny
and 10 cents per, like, we don't know.
They'll get that cost down and they will monetize it,
but they'll probably do like what they did with DALL-E.
So one, they're already making money
'cause they charge for the API
for these companies to build on it.
So there's money revenue being generated.
I think I had heard them like a 40-million-a-year company
right now, like annual revenue,
but their valuation is like 500X their revenue.
So they're making money,
but that their mission isn't to build marketing tools.
Like, again, I don't wanna like take us down rabbit holes,
but like OpenAI exists to build general intelligence.
These tools are just a vehicle for them to get there.
So to give the AI the ability to understand human language
and generate to see the world and generate the world
around it, like that's why AI exists.
So they're gonna find ways to make money,
whether it's through licensing.
- And I would gladly pay obviously for access.
- Oh, 100%.
- Let's talk about prompts.
We mentioned earlier that prompts are really important.
Let's give some tips to people
on how to create maybe better prompts, right?
Because like you said, you could just go in and say,
write me an email about a sale ending on Friday
and then we'll go ahead and do it, right?
But what are some of the things that you recommend
people that wanna go a little deeper with their prompts
might wanna consider doing.
- More descriptive, I mean, obviously,
and we kinda touched on that,
but I think what I would do is like,
let's say you had an example of, I wanna,
write this blog post, I wanna write these social shares.
I would test it yourself.
Like I would go see, write me a tweet about this topic,
see what it generates.
Okay, now write me a tweet about this topic
that includes this phrase in it or includes these.
Now, write me a tweet about this.
Like, just keep building on it.
- Yeah, and it learns we should say.
With ChatGPT, as long as you keep that thread open,
it'll remember all the stuff it just learned, right?
- Yes, yeah, to an extent until it gets
to kind of its limits of what it's able to remember.
But prompting right now,
your ability to prompt is very important.
It probably won't be as important in the not too distant
future because the next generation of these things are going
to understand the intent of the prompts.
What's gonna happen is,
us, the average user will put a prompt in,
there will be a magical AI layer
that improves your prompt, unbeknownst to you,
that, magically improved AI prompt
is actually what then generates the output.
So you may get an enhanced output
by being not very good at prompting.
It's gonna take average prompts
and make them good to great,
and you're not even gonna know it's happening.
- That's just crazy, I mean,
to think about that because I can also imagine
my average emails being improved
when I send them to my boss.
- Yeah, and that you're seeing tools like that pop up.
So what's happening is people have access to these APIs.
They can conceive of these things
like a YouTube summarizer or an email generator
that learns from my emails and they're building one-off apps
many times it's just like for fun on weekends,
not even building companies around 'em yet.
They're just developing these tools to see what's possible
and then they'll like make it available
as a Chrome extension or something.
So the explosion of generative AI tools
is going to be really hard to keep up with.
- A couple of tips that I've learned
from having experimented pretty extensively with ChatGPT is,
it tends to, first of all, it has almost perfect grammar,
but it tends to be somewhat repetitive
and it also tends to be very structured.
So what you can tell it to do is say in a casual tone.
I found that if you add any casual tone to the end,
it'll actually generate a more
easy, less computer-sounding generated thing.
You can also tell it to add emojis
and also to add animated GIFs.
Now, it won't actually animate the GIF,
but it'll give you a prompt to be able
to go over to Giphy to find the GIF.
So for example, it'll say GIF of a rotating clock.
You know what I mean?
And then you go into Giphy
and you type in GIF of rotating clock
and then we'll actually and it'll tell you
exactly where it recommends that you added in there.
I also think it's really cool that you can
kind of use, for lack of better words,
Boolean (indistinct) logic in it in a casual sense.
Like you can say without mentioning this.
So if you seem to notice that it's that, like for example,
it seems to be over-emphasizing the author
or something that you're talking about,
you can say without mentioning the name X, Y, Z, right?
I also really love the idea of flipping
between first and third person, right?
So if you want it to be author in your name,
it might actually write it as if it's you.
I'm excited to come to you today with an announcement
versus write it in third person.
It will flip that automatically.
My guess is you could also put copy in there and say,
write this in first person if it's written in third person.
And these are the kinda things
that normally you would have to communicate to a writer
with to be able to get this stuff figured out.
I've also found that you can get really nuanced with stuff.
For example, write about a ball
that happens to be blue that's flying through the air
and is gonna be caught by a person
in the outfield with a glove on.
You know what I mean?
Like the more details you give it,
the more it'll understand that you're talking
about something that is happening in a baseball field.
I don't know if you've discovered any other little tips.
- Yeah, so simplifying is a nice one.
Like write it as a seventh grade level.
Like, okay, now write it in this one.
The other one that I think most people
don't realize is add strategy to it.
Okay, now write this for small businesses
in financial services.
- And it'll actually adapt it and you'll be,
like, damn, that's pretty good.
And so I think challenge it to verticalize your content,
to like drill into specific personas.
It's surprisingly good at strategy.
- Wow, well, Paul,
I can assure you that everyone listening and watching
has their mind exploded as a result
of this conversation in a good way.
So tell everybody where they can find you,
if you have a preferred social platform,
mention your course that you've got working on or whatever.
Yeah, and then wherever else they can find you.
- Yeah, so marketinginstitute.com
is the home base for all of this.
We have our education, piloting AI for marketers
is a kind of step-by-step learning journey.
17 courses we've developed.
That's a great way to understand and apply AI.
We have our Marketing AI Conference.
We have an AI Writer Summit.
So it's basically events and education
and content and it's all living there.
And then our "Marketing Artificial Intelligence" book
can be found there as well.
Personally, I'm very active on Twitter
which is @Paul Roetzer and LinkedIn,
and definitely reach out.
Tell me you heard this podcast.
I love like connecting with listeners and finding out
what was valuable and interesting to them.
It helps me.
I of think myself as a storyteller.
Like I'm just trying to help people understand this stuff.
So if I know it resonates with people,
it helps me help more people.
So yeah, I love to connect and hear from people.
- Yeah, and folks that are listening,
Roetzer is not spelled like it's sound, it's R-O-E-T-Z-E-R.
We both happen to have Zs in our last name,
which is kinda cool.
Paul, thank you so much for coming on the show
and sharing all your insights with us.
We're way better because of it.
- Oh, I always love doing it, man.
It's great to talk with you again.