[MUSIC PLAYING] Hi, guys.

I'm a little nervous.

I work at home by myself in my pajamas most of the time.

So I'm like, wow, there are a lot of people with bodies and nine little screens.

All right.

I'm here to show you a little taste of what I have done personally with AI over the last year.

One small piece of it.

I probably made almost 200 different things, but I just picked a couple as examples.

One thing I want to make really clear, I am not a developer.

I always wanted to be a developer, but it seemed like strange magic.

I'm not a tech guru, a machine learning researcher, anything like that.

I am a creative thinker with too many ideas, too few hours in the day, and too much pride to ask for help.

Everything that I'm going to show you I did by myself at my dining room table with tools that cost combined a total of $70 a month.

I don't use AI to do what I do, but faster.

I think that is really important.

I'm sort of with Fred, where I'm like, push a button, get a whole campaign.

That's not what I'm interested in.

Of course, I'm a creative.

I'm like, no, I don't like that.

I work with AI to do what I couldn't have done before.

Unlike Fred, I loved AI from the second I touched it.

And I'll give you an example of why.

A couple of years ago, I was reading Little House on the Prairie to my daughter.

And there's a scene where they're traveling across, like in Kansas, I don't know where.

But they're in their wagon, and Lori Ingalls Wilder writes, "There was nothing ahead of me but the waving grasses of the prairie.

And there's nothing behind us but our own wagon tracks."

I put my kid to bed, and I went downstairs, and I cried because I was like, there's nothing else to discover.

I'm never going to have that feeling.

I'm never going to have that feeling of being by myself in front of endless vistas with no marks behind me but my own.

And then I touched GPT, and I was like, it's fucking on.

So like I said, I'm going to take you through some of the things that I've made, talk a little bit about why I'm sharing those things, and try to get this all done without going over.

Speaking of things I couldn't have done before, the very first thing I ever made in mid-journey was a series of vintage photos of robots in hot tubs.

You can't put a robot in a hot tub unless you can.

Then I made-- I love learning.

I will learn about anything, anytime.

I always say that curiosity is my superpower.

So I made an endless curiosity engine, which you just say, tell me something new, and it tells you something, and then tells you three more things that you can learn about.

So here, I think it was telling me about octopi.

It really likes to start with octopi.

And then it let me say I could explore animal superpowers, alien life forms on Earth, or the science of blood.

My daughter loves this.

I love-- imagine if schools, instead of blocking chat GPT, actually let children have things like this, where they could follow their own curiosity and learn anything.

Then I kind of realized, I was like, oh, shit.

I can make these little applications that run inside a chat.

And that's when I started to have a lot of fun.

I made a multiverse decision maker.

I find the multiverse really comforting, because I'm like, OK, well, there's a timeline where I made every possible decision I could make.

So I don't have to get it right.

I just have to decide which timeline I want to live in in this version of me.

So the multiverse decision maker, you say, here is a decision I'm trying to make.

Here are the options as I see them.

Then it says, here are two or three other options you haven't considered.

And then it plays forward all of those futures and writes a little immersive scene of you in that future so that you can see what it might feel like, and makes it a little bit easier to make a decision.

I love language models.

I think they're interesting.

I think they're fascinating.

So I just, a few weeks ago, launched a newsletter that's written entirely by large language models.

So Claude, Gemini Ultra, and GPT-4, their only job is to go discover something and send back a transmission.

Yesterday's transmission, Claude was wondering if it needed a body to be truly intelligent.

Gemini discovered the hero's journey and said, I wonder if I could have my own hero's journey.

And then GPT-4 taught me about fermentation and said, have we left aside too many of the ancient ways that have something to teach us now?

So it's been really fun watching them have discoveries.

Now we're going to get a little bit into the entertainment portion.

I made a website called needsmoreboom.com that will rewrite any scene into a Michael Bay scene.

So think, one of my favorites is the scene in "Lion King" where Mufasa dies because Scar always shows up with a leather jacket and aviator glasses, riding like a robotic wildebeest with a rocket launcher.

The entire website, I built this in 10 hours.

I did all of the front end, all the back end, wired up the API.

And I would say a good half the time I spent building this was getting the prompt right.

Also, I wanted to show this because you don't have to build everything inside ChatGPT.

But I made this.

People liked it.

And it cost me $1,000 in API costs in one week.

And I said, you know what, I'm building everything as custom GPTs from now on.

And so this custom GPT is called TGIF.

And it takes a moment in your life and turns it into a '90s sitcom.

So you say, here's the moment.

It asks you if there's anybody in your life.

Then it writes a full episode script and generates a promo image.

This is a "I killed a fish last week."

And that was the prompt I gave in for that.

I wanted to see if I could connect with APIs, if I could get something to happen inside ChatGPT that lived outside.

So I made the Spotify Storyteller.

You give it a title.

And then it writes a story that is just a playlist that it publishes to Spotify.

So the title I gave this one was "The Vegetables in Your Fridge Are Unhappy About Going Bad."

And so it says here, "The Unhappy Vegetables, a narrative playlist capturing the journey of vegetables as they go from being fresh to their eventual decay, told from their perspective.

Fresh in the dark, decaying silo, V2, wasted summers, throw it away, crying alone."

OK, this one's really fucking weird.

This is Operation Character Clinic, where you can play the game Operation inside ChatGPT on famous characters.

I wanted to see if I could get around some of their blocks.

So here you see I'm doing brain surgery on Mary Poppins on the top.

I've given Bart Simpson a bionic arm.

And then my operation on Captain Crunch went totally awry.

And he exploded into a bunch of crunches.

My daughter likes to make-- does anybody-- if you have kids, you know what a Perler bead is.

If you don't, you might not.

My daughter loves to make things with Perler beads.

And she was taking-- like, looking at YouTube videos to figure out what to make, which really fucking bothered me.

I'm like, come up with your own ideas.

So I made a Perler bead pattern maker in GPT that you tell it what you want.

It uses the Python tool to generate a pattern that you can use.

So you can use it to make-- she used it to make our dog.

I used it to make an Adderall XR 20 milligram capsule.

And a Snoop Dogg-- Snoop Dogg standing next to something that's sort of solo stove-ish.

I made a tool that you can turn any day into a party.

You just say, I want to party today.

And it gives you a party theme, name, decoration ideas, food ideas, drink ideas, activity ideas, a playlist, and makes you an invitation.

Did it today is in Belgium.

It's a celebration of the iris.

So we're going to have an iris bloom bash.

We're going to do floral arrangement workshops at the photo booth.

And we're going to paint irises.

And then you'll notice that our playlist is all sort of floral-themed.

It's super awesome.

On a Saturday, if you have a kid, and you're so tempted just to let them watch TV, you just go, you want to have a party?

And then you push a button, and then you make up some stupid party, and it's very fun.

CleanPoPilot for people who have ADHD.

I was interested in how do I use voice mode.

So I plug this thing in, and I'm like, I don't know what to do.

And it just tells me what to do.

So you see here, I'm saying, I got five minutes.

She's like, what do you want to do?

I'm like, I want to load the dishwasher.

And she tells me how.

And she's like, how's it going?

I'm like, pretty good.

I got a bunch of stuff in the sink.

I'm opening the dishwasher.

She tells me, and then I say, I have a question.

Why is it so much harder to unload the dishwasher than it is to load the dishwasher?

And so this is a transcript of an actual voice conversation that I was having.

And it was really funny, because when I did this, I saw so many moments where without my CleaningPoPilot, I would have gotten distracted and started reorganizing my sock drawer.

And so I'd be like, I'm starting to think about reorganizing my sock drawer.

And the model would be like, well, that's a great idea.

Why don't we table that?

Or I'd be like, hey, I'm downstairs.

Should I take this thing upstairs, or should I put it on the stairs so that I don't get distracted?

The model was like, put it on the stairs.

I'm like, great idea.

Plant advice, but from your plants.

I'm really bad at taking care of my plants.

They always die.

And so I thought maybe guilt would make me more effective.

So this one, you upload a picture of your plant.

I also want to play with GPT Vision, so it gave me a reason.

I upload a picture of my plant, and then it tells you how it's doing in the voice of your plant.

And the plants are very dramatic.

Do you hear my leaves wrestling a melancholic symphony?

It is the sound of longing, a plea for the quenching waters that once cascaded through my soil.

For the tender touch that would remove the dust from my weary visage.

If anybody knows anybody at Miracle-Gro, hit me up.

I made Task Slayer.

You put in the things that you have to do.

These are all happening inside ChatGPT, by the way, the image generation, everything.

I made Task Slayer.

You put in the things that you have to do.

It orders them from easiest to hardest, and then generates an image for each one of it as a monster.

And then when you say it's done, it shows you an image of that monster being slain.

So you can see here that I had to take out the trash.

I had to fold the laundry.

Oh, and then if you don't get all your tasks done, somebody's pissed at you, you can talk to Amy DeZoole to know if you're an asshole.

I was curious about training data, because we know that there's a lot of training data in there, and I know a lot of it's from Reddit.

And so I said, can I get it to sound like an actual Redditor, and can I get it to actually have an opinion?

Because if anybody has worked with these models, you know they don't have opinions very often.

So Amy says, no, you're not the asshole.

Or she'll say, yes, you are the asshole.

If you want to know clinically if you're an asshole, I made a generative Rorschach test, where a Chad GPT generates the images, gives you the test, and then tells you what it discovered about you.

Mine, it says, the imaginative and perceptive responses point to a creative and emotionally versatile individual.

They have potential strengths in creative fields and interpersonal interactions.

And I was like, yes.

I made Project Runway GPT edition.

You upload a picture of yourself, and the judges of Project Runway tell you if your outfit looks good or not.

It was so much work to get Michael Kors to be mean.

He said, those pants need a rescue mission.

We're going for casual, not comatose.

I let them get way in on my outfit, and they didn't like my boots.

And I told them they were stupid.

Poster designs done entirely with generative AI.

I wanted to see if I could actually make an entire design, not just an image.

Looks like a can.

Oh, this is a fun one.

The real test of work that worked.

I said, you know what?

Forget can.

The real test is, does GPT have your work in their training data?

Was it represented enough that the model knows what it is?

So if you ask it about your campaign and it knows about it, it gives you a little award that says, GPT knows my campaign.

And if it doesn't, it gives you a participation trophy and goes, oh, but let me tell you about this campaign.

I do know.

And what's really cool is that trophies are generated every time.

It makes up the trophy, and you never know what you're going to get.

And for the creatives, one of the things I wanted to show is I use MidJourney in a really different way.

I'm a writer.

I don't want to say, make a thing that looks like this.

That's not fun.

So I take words, and I say, what art is hiding in these words?

And so the prompt that you see-- the line that you see here, the feedback that all of us have heard, is actually what went into MidJourney, along with some stylistic attributes.

Because I also like making collages in MidJourney because it feels less like stealing, because a collage is just fucking putting stuff together from other people anyway, so that feels better.

So this is just have fun with it.

This is make it work harder.

This is I'll know it when I see it, paired with my favorite.

I feel like I've seen it before.

And if you get really lucky, you get I don't hate it.

So all of these I've made for fun.

I won't go too much over.

God, this thing is stressing me out.

I'm going to go like two minutes.

Give me two minutes.

Thank you.

All of these things I made for fun to entertain myself in the nights after my nine-year-old daughter goes to bed.

I'm a single mother working from my dining room table in Durham, North Carolina, and I can do any goddamn thing.

So I'm not done yet.

I'm not done yet, everybody.

But wait, there's more.

I just launched my very first custom GPT with a national brand last month.

It was the wedding planning brand Zola.

They found that 85% of women feel like they do all of the work in planning their wedding.

And the men just show up and say, I'll help you with whatever you ask for, which is bullshit.

So this is called Split the Decisions.

It's a tool that helps both partners plan the wedding.

It asks your names.

It says when, where, how many people.

It asks each of you some questions about the wedding.

What are you nervous about?

What are you excited about?

What's your priority?

It asks a series of questions.

Which of you is most likely to actually enjoy negotiating with a vendor?

Which of you is most likely to really care about the aesthetics of the event?

And based on your answers, it takes all the wedding planning duties, divides it in half, assigns it to both of you, oh, and gives you a downloadable CSV with the task the person's assigned to.

You see my imaginary fiance Julia, myself, and gives you a link to an associated resource on zola.com to help you do that.

I made this by myself in two weeks with a single person on the client side.

We were originally going to do it as an actual product, but to get it in through the product was too much work.

It's too hard.

But if you can make something like this that you can test, that you can put out there, the costs of it are so much lower.

And then I do a lot of stuff with agencies that I can't show you.

I do a lot of internal tooling.

I made one agency a tool called the Art of OKRs, where every employee works on their OKRs with the help of Sun Tzu and Machiavelli.

I do a lot of bespoke-- I really love building personas of customers, of potential customers.

I use a lot of psychographics, like Tim was saying.

The program and language is psychology.

I should mention that I was trained as a therapist before I got into advertising.

I build psychographically-driven personas for customers that-- I don't want to brag, but do they answer in line with vast quantities of Simmons MRI data that I did not feed into them?

They sure do.

When you show them two pieces of work and you ask them which one they like better, do their answers correlate with what people say in real life?

They sure do.

But I do it because I want clients to remember that at the end of the day, you are not talking to 40 million impressions.

You are talking to one person 40 million times.

So yes, I make AI personas, but I make them to bring the humans that clients are trying to reach closer to the clients, to help them remember that they're talking to people.

Ideas are cheap.

Execution is valuable.

This is what I've been told since the day I started in advertising.

I've been doing this for almost 20 years, and I've been told over and over again, anybody can have an idea.

Ideas are easy.

Execution is hard.

Execution is valuable.

If you can't execute it, your idea is worth shit.

Sometimes people have been hard.

Not only is execution cheap, it's instant.

And my question is, what does that mean for ideas?

That's what I think.

If you want to start, you don't need one of these giant systems.

They're beautiful, but just start.

Like I said, ChadGBT is $20 a month, and I was able to do all of these things.

So if you don't know where to start, just start.

Sometimes the old adages hold true.

If you don't know, just do it.

And if you don't know where to start, you can always come talk to me.

Thank you.

[APPLAUSE] [MUSIC PLAYING] [MUSIC PLAYING] (upbeat music)

‍