AI-First Marketing With OpenAI | INBOUND 2024

Man, the Friday crowd, the Die Hards, I love it. My people, I'm Dane, I do strategic marketing at OpenAI. I'm really excited to be here at Inbound today. That video we just watched was actually completely generated with Sora. So we used a text prompt, we added an audio file at the end, we used a text prompt to create that video. So pretty amazing what we can already start doing with content today. Um, we've heard a lot of really big visions around AI this week. HubSpot did an amazing keynote on Wednesday where they talked all about this new agentic future. I'm gonna do a very different talk for you today. See, what I'm obsessed with with AI is AI Primitives, the tools you need to be successful to build from AI today. Well, I'm just as excited about the vision. I really wanna kind of have this session, walk away with, with things that you can use every day to really kind of uplevel your skills with working with AI. So hopefully you'll learn a few things. Hopefully you'll get some education out of this. My focus today will really be on AI Primitives and I actually want to start with a personal conversation that I had recently with a young gentleman, Dylan. So this is not actually Dylan, but I generated this dolly. It's a close approximation of. And Dylan is a 17 year old who lives in Northern California. He's very ambitious, he's into cross country, he loves to think about extracurricular activities, he's really into technology and him and I have conversations quite a bit about tech and AI in the future. And I was talking to Dylan and I was asking him, thinking about universities, what does he want to do? What does he want to do when he starts going and thinks about a career and shout out to the 5% of 17 year olds that know that answer. But for most of us it's a really difficult question. And Dylan kind of gives me this blank stare and I'm like, oh God. It's a tough question. And he kind of looks at me and he's like, he's like Dane. I'm like, yeah. He's like, it doesn't matter. I'm like, what do you, what do you mean it doesn't matter? He's like, he's like, I can do anything I want. I was like, what do you mean? He's like, well, I use AI every day. He's like, when I'm planning my cross country runs, I use it to kind of think about my fitness and my nutrition and how long I want to Run. You know, I'm studying my AP classes, I'm using AI to help me with the course curriculum. When I'm, you know, planning things I want to do on the weekend, I'm researching with AI and kind of funding fun things to do with my friends. He's like, he's like, whether I want to be a doctor or want to be an entrepreneur or a lawyer, work for a nonprofit, he's like, I have this super tool. He's like, I can essentially do whatever I want and I know where to get help and know how to solve problems. So like my career doesn't matter. He's like, I'll figure it out. He's like, I have AI that can help me. And I was like, wow. It's like this real profound answer. But what I loved about it was like their sense of optimism and like sense of excitement for the future. And he had this idea that no matter what was thrown at him, he can accomplish anything because he is known in these AI primitives. He's learned the tools and he's learned the foundations to solve any problem with AI. And that has made him incredibly empowered. And we're seeing this actually in the workforce today as well. Paul Graham tweeted this where he had spoken to a bunch of CEOs and what he's seeing is that, you know, basically 22 year old programmers are just as capable coming into the workforce as 28 year old programmers. This is kind of mind blowing, right? You know, someone with six years of experience, a 22 year old, is just as capable of them. Essentially what they do differently is they know how to use AI and they're applying AI to solve different problems in software engineering. And this is really important because, you know, we have a lot of marketers in this room. Our jobs are not getting easier as marketers. I created this list. This list could probably be much longer. These are things that I think about. But things in marketing are getting much harder, right? Like you are expected to, you know, deliver the same results with lower cost of acquisition. You have to have consistency across all your channels. There's a lot more channels out there. You're shipping things faster. You're doing that probably with less creative, creative and design resources. You have to operate in increasing regulations. You know, you have to go above the knot, the noise, the cringy, go viral. And you have to think about attribution roi. Like raise your hand if like five of these things are part of your job. Yeah, every hand just went up. So everyone is dealing with these things now. So it's no real surprise that when we actually looked at ChatGPT data and looked at enterprise data, the number one case for ChatGPT was marketers. We see more organizations deploying AI to marketers than any other role. So it's really exciting that marketers who need a lot of help are turning to AI. There's just one problem, it's how they're using AI. And I think you probably know where I'm going with this. There's basically one use case they're using AI for content marketing. That is a super use case for AI. This is a survey I saw recently and this really spoke to me where they've actually went out and they looked at, you know, how people use it daily, how they use it monthly, what have they tried and stopped, and what do they don't use it at all. And you can look at all the use cases are all essentially content writing, email writing, social copy, and a lot of things that may have started where they thought about design, you know, analytics, abm. They're not using that anymore for AI or they never given it a shot. And you know, this is totally not actually everyone's fault. You know, if you looked at some of the image creation and some of what's possible, you know, six 12 months ago, you know, there was a lot of room for AI to improve. But it's like really exciting. This technology is improving so rapidly and there's so new capabilities coming out all the time that's making this technology better. So what I really want to focus on today is what I'm calling the AI primitives for marketing, kind of this new sets of skills. What do you need to know to be able to do your job well in marketing? How do you approach AI? And you know, I've broken it down to a list of five, and that's research, data analysis, content generation, automation and coding and thinking. And really by mastering the foundations, understanding kind of first principles, how to apply these techniques to AI means you're gonna have that level of optimism and confidence that any problem you see in marketing, you will have a solution where AI can at least help you get started on it. So I want to start with research now. I think research is actually increasingly becoming important as a marketer. You know, OpenAI is obviously a research company. We really value research. But I look at the role of marketing today. You were asked to really understand your audiences. You need to know about your Personas, you need to know about what markets you should be into. You need to know all so much more detail and have this information and be able to access it quickly. Our traditional LLMs have not been great for research. They are trained on data up to a certain point and that's what you're accessing the LLM. I know if you do need research, you need real time data, you need real time results. The problem that, you know, I'm sure many of you face is that, you know, going online is not really great for research either. You may be clicking on a link, digging into information, trying to find what's actually credible and that's not really helpful. So we're really trying to reimagine what does it mean for everyone to do research and, and chatgpt and how every marketer can be a researcher. So we recently launched this prototype of SearchGPT. So this is our new experience that we really want people to start activating and thinking about research here. In this demo, you can see someone searching for global tax compliance, looking up real time tax trends. They can search and look at flight information, checking and seeing what's happening in Southwest flights. Even if they're planning an event, they can look and see what sort of exhibits are happening at the local event. So there's lots you can potentially do from here from research and this will unlock a bunch of new capabilities. Now I think it's really fun to show videos, but at OpenAI, we actually like to like show how our technology works in real time. So I'm going to do a real time demo of our new search GPT. So if we can switch to the screen here.

All right.

All right, cool. So I pulled up the search GPT prototype. So let's say in this scenario I am selling dental software and I want to expand to Europe. Is there anyone from Europe in the audience here? Awesome. We have a lot of Europeans. I love how international this event is. So I'm going to ask SearchGPT a really simple question.

What do I need to know about selling dental software in Germany? It's real fun typing in front of 7,000 people.

All right, so I asked you this question and you can immediately see through the search results. I'm getting a lot of helpful information. I'm understanding the marketing overview, understanding regulatory compliance, thinking a little bit about different trends, competitive landscape. So already I have a great baseline for information and research here. And what's really powerful about the research and the search experience is you can see the links, it's showing me the different sources of where this information is coming from. So if I actually want to click in and better understand One of the links. I have this ability to do this right here in research. So now I'm sure if you were actually doing research, you're probably trying to find a lot more information. You can engage and ask questions here. So let's kind of play this out. Let's pretend I've done this all. And now I want to understand how do I actually start thinking about marketing in Germany? And I know events are going to be fantastic for me. Everyone loves seeing our software. We want to go to events. Let me ask it. You know, really simple question.

Are, what are the best events for dental software in Germany?

All right, so now I'm going to actually research now, you know, I've been at Salesforce, I've been at Stripe, worked a lot of different companies. You know, traditionally, how I would have solved this question is I would have contacted a local agency. I would have had them do a bunch of research for me. They probably would have taken a few weeks to get back to me. It probably would have cost a lot of money. I would have gotten an event list. And this would have taken a lot of time. And this isn't how we want to do marketing today. We want to empower you to solve these problems with AI. And here you can see it's already pulled up a bunch of different events.

Put this in a table.

I prefer to look at my list in table views. So I'm going to ask it to quickly put this in a table for me. All right, great. So it's already laid out what events I should be thinking about going into. The German market. This one seems pretty good. The International Dental Show. So let me just click on this one.

I'll paste it here. Say, how do I apply?

All right, so now I'll ask a question. So I want to take action on this. It'll give me great details. What do I need to apply? Let's see if this works. I'll click on here. Great. You already can see the link comes up. Apply as exhibitor. So what I just did there in less than two minutes was understand the local German market. How am I thinking about building out my dental software company there? Understand which events are being prominent, understand the process for how I apply to them, and basically build all this research in just a matter of a couple minutes. So as you kind of think about how you're going to apply this today, it gives you a really great, powerful tool. All right, so we'll go back to our second primitive data analysis. Now raise your hand if you're someone where you see Data. And you get really excited and you see data and you're like, yes, this is going to make my day. All right, we obviously have a lot of, like, rev ops and growth and marketers here too, that really embrace data. So if we can go back to data analysis. Slide. Great. There we go. And I really want every marketer to have that reaction. I think I saw probably 25% of the hands come up in the room. Now, I personally, I came from product marketing, and I don't want to shame product marketers, but product marketers aren't known to be very data savvy. We're great at a lot of other things, not data. And when I became CMO of a company, I had to really understand the data, how our business worked, what was our customer journey looking like, and really kind of understand the core nuances. So I want every marketer to be really empowered when they see data and be able to kind of understand and get the nuances of it. So I want to show another demo of how you might think about addressing data. So if we go back to my screen here. All right, so in this example, I'm gonna upload a lead list from my computer. So this is pretty typical. And often I've been in meetings five minutes before I walk into a meeting. I might get a data set that has a lot of information.

So let me try this again. Fun of doing live demos.

Cool.

All right, that worked.

All right, so you might get some information like this, where you have a whole bunch of tables, a whole bunch of different channels. And often, as I found as a leader, it could be overwhelming. Like, I'm expected to look at all this data, understand the nuances going on, ask good questions, and ask insightful that help my team. And this can be really difficult. And where I find ChatGPT to be really excelling is vision. So I can really kind of drop.

This data in, say, what do I need to know about which channels are performing best?

Great. ChatGPT is also fantastic with typos, which I love. So here it's gonna take a look at the data, understand the context, and hopefully give me a response a second, where it's gonna share out kind of what the key insights it needs to think about for the data. All right, here it goes. So right now it's looking at leads generated by channel. It's understanding some of the relationships in the context. It's giving me cost per lead. So immediately I can go into this meeting and have a great understanding of what my channels are performing, where I'm seeing different issues, how essentially things are working and like have this context as well. So as a leader in an organization, this just gives you a whole new relationship with data. It allows you to make data feel very actionable, very insights. So we'll scroll down, taking a second to generate and soon it's gonna tell me kind of what are those top channels that I really want to focus up. So gives me ttoa because it looks like SEO is really doing really well. PPC and email are my top channels. All right, so I'll feel armed and know what to talk about in the conversation. But you want to go further with data. You want to think about how you might actually forecast and kind of build a strategy around this.

So let's say I have, I have 1 million to spend in 2025. How should I allocate my spend?

All right, so now you can start prompting questions right where we've asked the, the data way to look at it, kind of the key takeaways now we want it to help me kind of think about my strategy. And so here you can see it's starting to break out. Give me some good insights in terms of where I might be spending my context here and help me kind of understand where, where I should be going with my data. So as a leader, I've just done two great things. I've been able to kind of look, understand data really quickly and start kind of thinking about how I might build a strategy around data. So I want to do one more prompt here. And this is what I find actually to be a superpower of chatgpt is tell me what I'm missing. Identify blind spots for things I haven't thought about and areas I haven't dug into. It's often when we look at data, we are very biased by the information that we want to see. And I think this, what makes it really special is that we can help kind of generate the right content. So I'm going to ask one more.

Question, is what are the top three things I haven't looked at I should think about? All right, pause here.

All right, so let me, let me prompt this next question and see if this rolls.

It's running a little bit slow here.

Cool. So let's go pull this out. So again, I think this is a really great way that you should, no matter what you're doing is asking this kind of question, identify areas where you haven't really necessarily thought. And here you can see it's thinking about attribution, multi touch analysis, and it's giving Me context in terms of why it matters for my business, understanding what I'm doing too. And as you give the models more information, as you give it more context on your business, this will only be more powerful in helping you understand kind of core concepts and making sure that you feel really comfortable approaching data. So that's just kind of one quick example. All right, going back to the slides, the third example I want to talk about is content generation. Now, I bet you are all like, hey, didn't this guy just say that everyone's doing content generation? Why is he talking about content generation? Well, a lot has changed in content generation. You saw that video that we generated earlier with Sora, and it's good to have a little context here. When you look at what our models look like in 2023, all these models operated really independently of each other. We had language, image, voice, text to speech. But if you said something in speech, what the model was doing was actually converting that speech to text, understanding the context, converting that back to speech, and giving you the answer. That's a very high latency process. It's not that helpful. In May this year, we launched our GPT4.0, which is our multimodal capabilities. This allows for you to input any sort of content combination of text, image, speech, and have the model immediately synthesize it, understand it, and output in that same format. So this is really powerful, you know, as if anyone's seen our advanced voice capabilities. You can almost have real time conversations now with AI that can understand nuance, understand tone of voice, and kind of get that extra level of detail that really helps when you're thinking about creating content. We're seeing a lot of exciting applications of this today, so I want to do one more demo showing how all these multimodals will work. So in this example, I've basically created a text prompt and I don't know if anyone enjoyed the Olympics. I had a blast watching the Olympics this year, but I kept watching this beach volleyball and it had this beautiful Eiffel Tower next to beach volleyball. And I was just kind of wondering what must have been like in 1889 when they were creating the Eiffel Tower. So I wanted to do a video that would actually be reenacted of what it looked like in Paris in 1889 at the Exposition Universal when they were creating the Eiffel Tower. So let's see what it created. All right, so it's using Sora to create the video. All right. We may be having a few WI fi challenges. We were hooking this up in real time as we were going through this. All right.

Let me try one more time here.

All right. Unfortunately, I don't think the wifi is rendering here to create a video. But in this example, what I wanted to show was a video would be created and this would allow you to kind of really think about how you apply multimodals. So you could actually add text to the video, you could add a script to this as well, you could add your voiceover and really kind of create a whole new experience where you can actually customize and build content from scratch. So I think it's a really exciting vision for how we're actually going to rethink the framing and changing of technology with video. I would love to show this video, but it's not going to render. But it's a really exciting different way for content generation and lots of amazing companies that are going to be working on this today. So I want to go back to the next example, which is automation coding. Now, this is something that may not be obvious in terms of how you might apply this to your job in marketing, but one really important thing to think about is the cost of models depreciating and what does this open up for new use cases. So one thing you might not realize about AI is it's the fastest depreciation appreciate depreciating technology ever invented. The fastest depreciating technology ever invented. Our models used to cost $36 per 1 million tokens to run a little over a year ago. Now with our new Frontier models, GPT4 mini, our smaller model, that's only 25 cents per million tokens, we're seeing a host of new opportunities be created and a whole ways of adding AI into your products that weren't possible when the price wasn't feasible before. So I want to share a different example for how OpenAI is actually doing this at our company. So one of the challenges, and this is kind of a good challenge to have, is we get a lot of information coming in from lead forms. We actually have a text field, which I know most companies don't do because it can hurt your conversion, make it really difficult for you to understand it, and someone has to actually read all those responses. But what we decided was we wanted to use natural language to understand every single text field of every form that's being filled out on our website. So we actually can take a look at all these form fields, understand the context, understand the nuance of the conversation, and if it's a lead, we can actually score that Lead and route that right to the right SDR or ad. If it's something where it's a customer service inquirer, we can route that to the right customer service rep. If it's something where they want to know more of a biz dev or a product question, we can route that as well. So we can actually use AI to understand natural language, understand what's happening in our forms, and change the whole scoring and routing experience with the our company. Another example is coding. So I'm really excited about this as AI is really kind of a pair with the work that you're doing. As we look at opportunities to share your screen, you can actually see scenarios where you're coding and AI can actually be asked to look at the code, identify areas, identify ways to improve the code. And this creates a lot of opportunities. And I'm really excited about this kind of, this next frontier of just sharing your screen. In general, you can see this happening in design. If you want some information input on design, you can see it's happening in writing where AI can pair with you and help kind of inform how you're thinking about what you're doing and really give you real time feedback. All right, the last primitive I want to talk about is thinking. And this is perhaps the one I'm most excited about. I'll give you a little bit of a story in terms of how I commute to work. So I drive to work at about a 45 minute work commute to work. And what I'll do is I'll pull up my iPhone and I'll hook it up to my, my speakers on my, on my car and I'll take a look and see what meetings I have today. Am I speaking at an event like Inbound? And I'll start having me ask me questions, you know, what are key points I should be talking about identifying gaps in my knowledge, like sharing back different thoughts, helping quiz me do I really understand the fundamentals of what I'm talking about is this makes sense and AI is great at kind of having this conversation. We all know when we're brainstorming the power of that is interacting with someone. It's be able to talk to someone and kind of share and kind of beat up ideas and explore and thinking it's going to be a use case. I really hope all marketers start embracing and as we increase our context windows, we recently launched a feature called Memory. You'll be able to add more information into ChatGPT that you can work with you as a thinking partner. So Imagine taking books that you read and putting that in chatgpt conversations that you've had, you know, key reports, messages, all that context of a city of chat GPT. You can interact with it and use it as a thinking partner. So as we release our advanced voice mode, as you start interacting with AI, I really encourage you to not just, you know, use the tools, but help it kind of think better through how you want to go about execute on your strategies. All right, we are not the only ones thinking now AI. We just launched a new model, the first AI that can reason. So this is really exciting. This is our OpenAI01. What's really important is this is actually not part of our GPT lines models. This is kind of a whole new model together. Essentially. If you think about GPT4O, what it's really good at is really kind of predicting the next word. Can look at a lot of information and help you kind of build out what you want to do, whether it's content, image and so forth. But what it essentially does is it takes a snapshot of, of kind of what it sees and gives you that response. What we decided to do with O was to kind of like tell the model to chill for a second, to pause and think about a problem and come up with different hypotheses. So when you Launch a query in 01, essentially it's coming up with different ideas, different ways to solve them. It's critiquing itself and coming up with the best solution for how you might solve that problem. So we're asking it to think 20 to 30 seconds on a problem instead of giving you an instant answer. We rolled this out a week ago. We're really excited to see how companies are building with this. We're already learning a bunch of different use cases through this as well. I think kind of the key concept here though is thinking about tasks that you wanna do with AI in much longer time horizons. So, you know, while GPT3 was great at 5 second tasks, GPT 4.0 was great at 5 minute tasks, the new OpenAI one can start thinking about 5 hour tasks. We're starting to see really interesting applications of this in manufacturing and financial services and healthcare, where you have these long horizon tasks that you can ask AI to really think about. So those are kind of the five primitives I want everyone to think about. And I want to go back to that Dylan example. You know what excites me about Dylan is he really thinks that he can be optimistic and solve any problem in the future. And I hope like coming out of this talk. You've seen a few examples from this as well. But as marketers, I think we should have that same approach. Our jobs are not getting easier, we're being asked to do more. And AI can really help understand these primitives and kind of applying AI in different scenarios to make your jobs easier. And so as you think about research, data analysis, think about creating new content, thinking about automation and coding, and even thinking about thinking, AI can be a really great partner in how you develop these as well. And kind of my last, like, you know, call to action here is kind of just ask yourself to solve it. You know, we've looked at data and still something like 40% of the workforce is using AI on a weekly basis, where people under 20, that's closer to 75 to 80%. So I think it's really imperative, as you think about your next journey, where you want to go in the career, adapting these tools, being able to use these primitives in AI is going to make you have that confidence. Any problem that you face, you'll be able to have this tool that will help you solve it. So thank you for the time today. I really appreciate you guys and hope you all had a great week at Inbound.