Google’s Gemini 1.5 Pro Is BETTER Than OpenAI’s Sora… Here’s Why

Hey, we have a fire show for you today. We have big, big news in the AI video world. OpenAI has released Sora, their text to video model. Google has released Gemini 1.5. And one of the best parts of Gemini 1.5 is its multimodal, especially image to video search capability. We're going to break down both of these, what it means for your business. We're going to show some great examples.

And we're going to give you some.

Ideas of how to use them as they roll out. Access everyone. That and so much more on marketing against the grain.

Hey, hey.

We're coming in hot today. We are Talking about videos ChatGPT moment. We've got some major releases from OpenAI and Google Gemini that are going to change the face of AI, especially in video. I'm your host Kip Bodnar, CMO at HubSpot. I'm joined by my co host Kieran Flanagan, CMO over at Zapier and Kieran. We were just living a normal day yesterday and there and Sam Altman decided to pick another fight with Google. You know, like that's. We were just having a normal, having a normal Thursday and things got out of control.

But he's not trying to raise $7 trillion. He's launching things on the Thursday to try to steal the press from Google's Gemini 1.5 Pro launch, which is we are going to cover for the most.

Part, we're going to cover it, we're.

Going to cover a lot of Sora, but we will cover a little bit of Gemini 1.5 because it is an impressive release. But I wanted to kick off with a quick cool GIF and hand the flow. I love GIFs because you talked a little bit on a Tweet yesterday around AI's launch strategy, which I think is like, I love the kind of aggression between these different companies. And this kind of describes a little bit what's happening, right? Like every time Google tries to launch something, here comes OpenAI with a more impressive launch and steals the entire PR and kind of Internet attention away from Google. So why don't you go maybe describe.

What you covered in your tweets for folks tuning in, this is the scene. We're recording this show on a Friday. We're probably going to watch it on Tuesday, a couple days after this all happened. And what happened is that Google released Gemini 1.5 and we were all really excited. We were like, whoa, this is awesome. It's a great multimodal model. It has a 10 million character context window that's amazing. And that was the morning. And then in the afternoon, Sam Altman said, hmm, Google, you had a big announcement. Well, I'm going to one up your announcement because that's what I always do. He has a history of doing this and he announces Sora, which is their OpenAI's text to video model. We're talking about that in a second. But basically what happens is Sam's strategy, which I tweeted about, which is like, Google makes a big announcement, he just waits around. I am convinced that Sam just has stuff in his back pocket, but he's like, oh, you know, we don't have to release this this week. We could wait until Google tries to release something and Google does their release that same day. He has a much bigger release that upstages Google. The entire Internet is losing their minds over Sora. And we're going to talk to you more about why in a second. Then what did Sam do? He took to Twitter and just asked for people because Sora is not available for people to use widely yet. And so he was asking for people to tweet him prompts and then he was tweeting back the videos of those prompts, basically doing real time demos of this new model.

This is doing part of why they win. This is part of why they win, right? Because Sam Altman has realized that the Elon Musk approach to being a founder is truly great marketing. And Google are still launching things through a traditional large corporate way. And Sam Altman is launching things in a modern Internet way.

He's like, I'm just going to hang out on Twitter and launch stuff in real time for two weeks, for two hours and just make it happen. And so if you're a marketer, you're a CEO, think about this. It's going to have, you're going to have to shift your strategy more co led, more aggressive, because the world is getting more aggressive on the marketing front. And then I was kind of thinking of this, Kieran. And then Sam has the audacity to post this tweet, which is like, OpenAI is the most talented and nicest group of people I've ever seen in one place, working hard on the hardest, most interesting and most important problems. And basically he's like, you should come work with us. So after he does this live demo gets all the hype, he's like, yeah, you've been following me. Don't you want to come work with me again? And so then he gets all these people excited about working at OpenAI and, and then those People come build even better products so that they could show up Google again in the future. We're going to have a whole year of Sam just showing up Google. Like, I don't see any other path forward here.

Google are taking some actual interesting shots at SAM and OpenAI actually as well. So in their Gemini 1.5 release paper, in Gemini's 1.5, they have a big release paper. I think it's like 30, 40, 50 pages long. And so I watched multiple videos on it. There's two, like, kind of interesting Easter eggs in there. One of them is like really explicit to OpenAI where they show a graph of Gemini 1.5 and Comparable to OpenAI in terms of the amount of tokens or something it can take in through the API and basically just shows you like, we are much better than Google through the API usage. But the one that's actually more interesting is, I don't know, the backstory of Sam Altman being apparently, quote, unquote, let go by Paul Graham at Y Combinator. Whether that is true or not, I think Sam debates that. But they have like this one of their examples they show in terms of how Gemini is really powerful for context Windows is they trained it on all of the papers of Paul Graham. Now it's interesting they picked Paul Graham. Right? Like, you could say, like, there's some like, interest in Easter eggs in that paper in terms of like trolling each other back and forth.

Well, I got a couple things. First of all, Valentine's week, February 24th, I think goes down as Alpha Dog CEO week. You had Zuckerberg come out and basically try to take down Vision Pro. You had Sam trying to one up Google. Elon's doing his Elon thing. Very, very like high testosterone week this past week. And you know what, while the Google trolling is good, Kieran, I want to show you my favorite trolling that comes from the OpenAI side of things. And it is literally from the Sora product page. The Sora product page. And if you look right at the top, it says all videos on this page were generated directly by the Sora. By Sora without modification.

Yeah.

And that is just a complete dig. The fact that Google completely faked a lot of their last Gemini videos for their big, big keynote.

I love this. This is fun. This is fun marketing.

Like, it's good marketing. It's a shot at Google and exactly what you need the person reading this page to understand.

Right.

It's actually a great product page.

Right.

This is a fantastic. If you're just like hey, what's good product marketing look like? This is a good example of product marketing. Clear value prop. They got the three reasons like the three components here, capabilities, what can it actually do, how they care about safety and the research behind it. And like th. Those are the three things you're going to get on this page. And it's like all in the header. You actually have a background video of a Sora image right in the header. Really good page, right?

It does show you that arguably at the moment OpenAI just have the public's affection in a way that Google don't. Because Gemini 1.5 is arguably a comparable if not better launch than Sora. I know that's kind of probably controversial, but if you actually.

Oh, now that is a, that is a take.

If you go through the research paper, it is objectively a better model than GPT4. You can actually look at all the comparables and text, audio, video, it's comparable if not much better than GPT4 in a bunch of different areas. Better context, better on a lot of things. It's also multimodal from the ground up. We've talked about that from, from way back that they built multimodal in the ground up. So some of their demo videos are very impressive. But again they, they are not being launched in the way that OpenAI launched. OpenAI have clearly give the model to a ton of influencers and designers and people who are really have really large audiences on Twitter. Like their launch is just more prevalent across the entirety of the Internet. And Google just struggled to get the same sort of traction for their launches. And I think that is like a really interesting juxtapose of what's happening where you know, you have one brand that is kind of using modern content marketing, Internet, you know, Internet awareness tactics and Google are still kind of using the same sort of corporate way to launch products. But Google Gemini 1.5 is a huge leap from where they have been. Like I have to give credit toward the, to the progress that Google are.

Making, but both, both huge announcements. I want to dive a little bit more into the Sora announcement in just a second, but I have to, I'd have to say we do have to predict and call one thing that the fall of OpenAI is going to be one of the most elaborate falls in history. Like in terms of public sentiment, I think the company is going to do great. But the public loves to build things up and then tear them down, right? And OpenAI is on that build up path. And when that tear down happens, it's going to be ugly, which is why they are so focused on safety. It's why they're trying to be pretty open with their research. Right. They're trying to prevent that safety steep fall from grace in public opinion. And you're seeing it kind of happen in real time. So as you're watching all these things come out, also keep in mind kind of the, the drama, the story of what's happening here.

So this is like buried in the technical paper. Sony dorks like me read it. But like, this is what I was saying. If you look at some of the charts, like, it's pretty, you know, it's not, it's not being secretive or like, oh, like we're better than no named. It's like, no, we're much better than OpenAI. Now the thing is, OpenAI are clear that they only accept 128k. So everything after that you would expect errors because it's not part of the functionality. But Google are calling out like, we are much better. This is how well it retrieves. And so the red ones are the errors. But Google's much better. But this is the one I thought was like, really interesting. Right. Like following the, like, they used all of the essays of Paul, Paul Graham here as well to show some of the advantages that Gemini Pro has over OpenAI. So yeah, I love this back and forth. I think it's super cool and it's forcing companies to launch faster, do great work.

I think one of the core takeaways if you're a marketer out there is that competitive marketing is getting more aggressive and more head to head just in the world. Right. And like that was, what you just showed was a great example of that. And if you're thinking about where to balance and lean on as you're doing competitive comparisons and everything seems like it's trending a little bit more aggressive right now.

Right, right.

And that's. And I think that that is just because there's just more transparency and the buyers are appreciating that kind of direct honesty more and more.

Right. Okay. Do you want to show some videos?

Yeah, yeah, we got, we got to show some videos. All right, I want to show you. I want to show you my favorite video. I watched a lot of SORA videos, a lot, and this is my favorite. Bill Pebbles, who works at OpenAI. He created a whole zoo video of a video about different animals in a zoo and he called it Bling Zoo. So it's animals with jewelry. Like this turtle is iced out. Just diamond out this money Monkeys wearing a crayon, right? It's. And it's Bling Zoo. It is hilarious. And it looks incredible like that. Like the turtle, the tiger. Like these animals look freaking incredible. Like, how did you enter a couple sentences and get this. That is pretty mind blowing.

And do you know what's actually really mind blowing? Like, do you remember where we were a year ago? Let me just show this.

Yes, please show people where we were a year ago because I'm gonna stop sharing because it is very important, right?

This is the side by side comparable. Let's actually show the side by side comparable. Okay? This is under a year. Everyone remembers the Will Smith eating spaghetti. That was a year ago, right? Like when you actually look at the side by side comparable. And this is like the model today. And this was the prompt here was to have a woman describe the outfit walking down Tokyo Strip with like some neon lights in the background, right? And this on the right hand side, you could start to see it being part of a movie intro or an ad intro on the left hand side, you would never think to yourself, wow, like at some point I can start to use this Will Smith eating spaghetti video in any kind of meaningful way, right? So the, the evolution of this in such a short amount of time is the thing that I think is both amazing and definitely a little scary because it's just becoming so realistic so fast. One of the other ones I quickly wanted to touch on as well, because I think there's a real business here. It comes from Sam Altman because I was on his feed last night. Because as you said, the cool thing he's doing is just taking requests and then creating videos for people. But this one here, let's get it. We can get him bigger. This one here is really interesting, right? Because the prompt is an instructional cooking session for homemade gnocchi, right? Hosted by a grandmother social media influencer. And it had the set. And you can imagine, right? You can just create your own video.

Italian grandma influencer recipes.

Yeah. And for sure you can create your own recipes. You can have them taught to you by AI. You can have a character like you.

Can imagine the backlash by some from some dude in Florida pretending to be an Italian grandmother is going to be hilarious.

But if you are hellofresh, right? Would you not. Oh yeah.

If you're a brand, go and crush it.

I would integrate hello this into hellofresh straight away. I'm a hellofresh customer. I really enjoy it. The thing that would make it much better is having videos inclusive. Like if every recipe had a QR code and the QR code was an instructional video that was on YouTube that the actual unit economics to do that is too high. Right? It would. Every single recipe to create a video for is much too high. Not when you can have AI do it for you. Right? AI can both create the recipe and then create the video for you. Now the only thing is I will say the thing that you can stall do to tell the difference between a kind of AI video and a real video. It's still within this. What do you think it is when you see this video?

Hands.

It's the hands.

It's always the hands. It's the hands are really hard. The ones are the giveaway baby. The thing we got the hands.

The thing that is going to save the humans race is being able to like see the funniness in the hands. But that's. I don't know if you have any other ones. I thought that was. That was one of my favorite ones because I instantly saw it and thought why there's an entire business you could create with this overnight.

That was one of my favorites. I want to show one other because I think it's a different use case. That's going to be really interesting. Right. We've talked about this a little bit in the past, but this is an example of how you could use a model like Sora to animate still images. So like on the. On the left here you have the still image and on the right you have the AI animated version of that. So imagine taking photos, different things and taking them and making whole stories, whole videos, whole movies that are based on like some picture you were able to capture out in the ocean or in a forest or in some type of environment. Right. Like that's pretty mind blowing.

This is the multimodal aspect, right. That people still think through. Wow. I can go and create a video through text. You can actually add a image and create multitude of different videos from that image. You can out of it. There's a really great example of someone who uploaded a video and then created multiple different iterations of that video through the multimodal. The other one that I started to realize is going to be huge. And I don't think it really dawned on me until I saw some of these examples is video games like AI is.

Video games are going to be huge. This is an awesome example.

This is a really good example. These. The prompt was drones navigate in around Mars. But it looks much more like a video game than like anything, you know, kind of a realistic video. But I think AI is going to be able to create whole immersive worlds. And again, this is the content for one that we covered some time back where over the evolution of AI, I suspect things are going to get really weird, especially for marketers because people are so weird, man. Create their own media for themselves. So how do you ever market anything to them? Like I don't even know, like they have their own YouTube person that they've created. Yeah.

How you actually market to people in the future is going to be fascinating.

So fascinating.

Well, okay, so we've played around, we've showed everybody, we've talked about the capabilities of Sora. The challenge with Sora is that it's not out for any of us to use yet. And I think it's going to kind of be slowly rolled out because something we'll talk about a minute. It does take a lot of compute resources, kind of compute constraint in rolling this out. Before we kind of finish off on Swara and we finish off on Gemini and everything for today's show, what do you think the core early use cases of really good text to video is? Because I want to go and do a little pat on the back. Karen, in our prediction show, one of the things we predicted is that 2024 was going to be the year of text to video. That AI video was going to have a massive breakout. Here we are six weeks into the year and it's here like it is having a massive, massive breakout. What do you think some of the early use cases, if you were somebody and you were getting an early access to Sora over the next few weeks or months, what would you be doing?

It means that those stock videographer sites are obsolete. Right. So there's a lot of marketing material. People make a lot of things that companies make where they go to these kind of stock video sites and get these snippets and integrate it into an overall video or integrate it over in an overall marketing campaign. And so I think it's at that level. Right. It's not at the kind of I can custom create a entire creative ad. I could be wrong, but I think the first iteration is not there. But I think these kind of like the things that you're doing today where you're using this videos from stock video library, you can actually get much better results just using Sora. The other one that I think is really interesting is instructional videos. That's why the cooking one really resonated to me is, wow, I can actually create in these kind of educational videos to go with all of the text content that I'm creating. That to me is another early use.

Case that might be like I want to, like I want to talk about that just for a second because like we rolled out like a big update to the internal version of HubSpot CRM that we use at HubSpot to all of our sales team last week. Kieran like a whole revamped of how they do prospecting and a lot of their workflow. It was amazing. And we rolled it out using heyging with AI Alex, which was just an avatar we created in.

Hey Jin.

That did like a perfect. The best possible version of like a three minute training video for a sales rep.

Exactly right.

And so what you're saying is as AI video gets better and better, those use cases just become ubiquitous and there are so many of them. Think of all the things we use loom for. Think of all the things we use zoom recordings for. Even a lot of the text based education. Like you said that we're going to take into short form video and even mid form video is going to be pretty amazing.

Right? And when we get into Gemini 1.5 Pro, I have a kind of take where I think that this is the reason, part of the reason that YouTube is likely going to become the premier search engine in the future versus like a Google or anything else.

The other quick thing, hold on that, you know, we talked a bunch of listener feedbacks and one of the negative feedbacks they got, we got was like you all try to make us watch on YouTube too much. Sometimes we just want to listen to the audio. It's like there's a reason we want the YouTube. The YouTube is where we believe the future is. Doesn't mean we're ever going to abandon the audio. But we are going to be all in on the future of video and the YouTube is always going to be the best experience because like you and I just have deep conviction that's that's the future.

That is the future. Like this is multimodal, right? You have to start to learn how to create things in multiple formats. And we have the audio and so you have the choice for the user. I just believe over time the user will start to prefer YouTube and part of it is actually the re like that. We're actually going to have this whole boom of video created around tons of educational topics and do it at scale. Where are they all going to live on YouTube. So YouTube is going to. I think today there's 3,600,000 videos uploaded to YouTube every single day. I can see that being like 6, 7, 8, 9, 10X. I don't know how much can Google but like you actually add AI could.

Easily be 10x, right? The complexity of the algorithm and discovery is going to completely change.

That is the thing that I want to cover on Gemini Pro. One thing I wanted to show actually real quick because I think this is a great video you cover. You talked about the base compute, right? So like what's the biggest barrier of these models getting better?

Yeah, like why aren't we all using Sora right now?

As why does the sound admin want to raise 7 trillion to build chips? Because he knows compute is the biggest barrier to actually reach an AGI. And you can see it in this cool video.

It's all, it's compute and it's energy, right? So it's like how do we get high, low cost energy and how do we get compute up?

So this shows the same video. So it starts with the base compute, right? This is a. This is like the Will Smith eating spaghetti, right? Can't really make out what it is.

Terrifying.

4X the compute. 4x compute is pretty good. Look at the 16x compute, right? Same thing. Only thing that changes is the compute. This is why a lot of people are starting to think that the only barrier to actually achieve an AGI is not in how the models are trained or some sort of breakthrough there, but actually just computing. Which is why we need 10% of the world, stupid.

Basically.

We need the world. We need 10% of the world GPT to start making these different chips.

You, you came out with the hot take at the beginning of this show, which I think was flu. Flew under the radar that you thought the Gemini announcement was more important than the SORA announcement. And this is Gemini 1.5. Why? Why is it more important, Kieran?

Like the token length is a million. Like you can actually train the model. You can add an incredible amount of data to, to your interact to improve your interactions. Right? But like one of the things that I really stuck with me is this here I want to show you really quickly. Sundars. So they have a 44 minute silent Sherlock. Old Sherlock Holmes video, right? And so the first, the first thing they did which was really interesting was they said find me the place in the video where someone takes a coin out of another person's pocket. That's what we. I skipped over and it got the exact point on the video. And then you can actually say tell me what, what that person was doing and I can give you the exact context of what was happening within that scene. But this here is even more unbelievable. So look here on the left hand side you see this sketch that they've uploaded here, the little white. So I'll play it from here. What is the timecode when this happens? This is the example, that's all they say, of a multimodal process where we combine text and image in our input. Now there's some latency which you can see the fast forwarded here like took about a minute to get this back. So they got the exact scene that's wild in the video from this sketch which is just like a, you know, a pretty basic drawn of like water falling on someone.

Pretty basic. Like that is the most basic drawing you could ever do, right? It's stick figure.

Think about this, right? YouTube Today, today is depending upon how you measure it. If you measure it based upon monthly active users is the second biggest search engine in the world. Now if you measure it based upon search processed, it's not because it's not a text based engine, it's a video based engine, but just I think in the future we will measure search and quote unquote search engines based upon monthly active users, not search volume process. Because you actually see a lot of these AI chat products moving to subscriptions. Like Gemini Pro itself is actually going to be a subscription, you get a free trial. And so the. So today YouTube is the second most popular search engine on the planet. And it's terrible at search, right, because it's video based. But there's just a whole vast amount of content hidden within YouTube which is why I think Google have a huge head start on OpenAI in terms of having better models. And I think gemini Pro unlocks YouTube as a search engine. If you combine that with the fact that we're going to be able to create educational video through their own avatars. Like everyone could be a YouTube influencer because I don't have to be on camera, I can have the AI be the influencer, I can have the AI create the instructional video, I can have the AI uploaded to YouTube. So YouTube actually grows exponentially in terms of the amount of content. Now that in and of itself would be a problematic if not combined with Gemini and Pro 1.5 which unlocks all of this content as a search engine. So you could, you can imagine I could say find me all of the places within all of the videos on YouTube that Steve Jobs talked about, how to do incredible product launches and then build me an app that has the playlist of those clips. Why would I ever go and do that in text again when I can just go and build.

Why are people ever going to search with text again? It should actually be the big take. One of the big takeaways of the Gemini 1.5 launch. Right.

So that like that to me is that the, the aha for me was how well the multimodal works which is the example I give you. But there was multitude of different examples where they just added sketches and they added different elements different like text based text, audio and video. Plus if you actually read the research research paper, it's better today than GPT4 and all of those different things. But to me the aha moment was like wow. It makes video searchable. It turns video into like a search commodity.

And the video search is dope.

Probably better than like most webpages. Right. Most of the content on YouTube is better than most traditional web pages.

Which, which is interesting. Right. Because if, if, if text to video creation takes off, video search has to get 10x better.

Right.

Right. Cause we're just going to have a multitude of video. Exactly. That we're not going to be able to figure out. And we're never gonna be able to properly label it and tag it with text and do all the crap we do now.

That's why I think those two launches combined were actually really interesting because SORA allows us to create video on mass and in particular like the educational video is where I think most people will start. And Gemini 1.5 Pro, it turns that into like a real commodity for Google. And so Google kind of win because like all of the OpenAI source stuff gets uploaded on the YouTube and YouTube get to commoditize that and turn it into like really valuable search content for users.

It's actually very complimentary. Like after this announcement you're like oh, OpenAI is leading the text to video race. Google is well positioned to keep owning the video search.

Right.

Race. And so all of those videos created an OpenAI. Google's gonna monetize with ads, you know, and that's, it's actually very interesting for both companies involved.

YouTube is the single greatest acquisition of all.

Oh, it's.

Someone would need to like tell me what's even close to being as good of an acquisition as that company.

Instagram and WhatsApp are the other two big ones. But YouTube is the best acquisition a company ever did. We'll have to do a whole feature show on that. Okay, what else? Is there anything else before we go to like. I feel like we, we covered a.

Lot of ground so we could end On Julia, who's coming on the show. Friend of the show.

Oh yeah, this is good.

She covered a couple of like What? So why OpenAI objectively are just a large company doing incredible work in terms of launches, right? Like how are they operating at such speed? And the speed we showed you, the great comparable is Will Smith. An odd version of Will Smith eating spaghetti in reverse. Like he's actually coming out as mad. Instead of going in to today a year later. A model that could probably fool most people. You can create videos and you would never know an AI created them. And so she, she talked, she kind of posted this, which was really Sam's take on how you build a company that operates at speed. And he talked about a couple of things here we can kind of just end in which is, you know, concentrate on a small number of bets with high conviction. I think you and I believe that stuff as well, right? It's better to not be spread too thin. Communicate clearly and concisely. This is. We've talked about this simple and repeat number one in my book, dude, I love this one. Fight bullshit and bureaucracy.

Every time you see it, get other people to fight it too. So good.

Do not ship your org chart. Do not let your org chart get in the way of doing great work. Outcomes are what count. Yes, yes, yes. Do not make good process. Fuck process it.

It's process does not matter.

All outcomes are what matter. Spend more time recruiting. I give you a lot of credit for this. I think this is one of the things you are exceptional in that you've always taught people in HubSpot hire talented people. Everything else, take care of yourself, of itself. And you.

You all details after that baby.

And then superstars are even more valuable than they seem. You and I have talked about this for a couple of people in HubSpot where we had to go to bat for them. And we were like, there's a big difference between someone who is this good and someone who is second. Like, the difference is just do anything you can to like go to bathroom.

It's not a linear curve. It's an exponential curve right between the very best people and even very good people people.

So I think all this to say that Sam is a really great CEO, an incredible leader and that company are shipping fast. But wow, they have some competition. They have some real competition and Google and these other open source models that are trying to catch up. But Sora is, I think another chatgpt moment for them and they keep getting these launches, right?

Exciting moments in AI. We are really six weeks into the year and exciting stuff is already happening as SORA rolls out, as we as we do early experiments with it, we will come back and do future shows. I'm sure you'll hear from us a lot about AI text to video this year. I think it's going to be one of the core themes of this year and it's going to be a lot of fun. We will see you all very soon on the next episode. Marketing it's the great this data is wrong every freaking time. Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated.

Whoa. I can see the client's whole history.

Calls, support, tickets, emails. And here's a task from 3 days.

Ago I totally missed HubSpot grow better.