OpenAI's HUGE Mistake: Sora Launch + Product Review #ai #openai

Everyone, we have been waiting for the big text to video release of the year, OpenAI Sora. And it's finally out and we're gonna tell you good. Is it bad? Maybe somewhere in between. Watch today's show, we're gonna break it all down for you. Let's get into today's episode. On Monday at 1pm Eastern Standard Time in the United States, OpenAI was supposed to release Sora about an hour and a half before Marquis Brownlee, one of the biggest tech YouTube personalities breaks the embargo, releases his whole 16 minute video of Sora on X on YouTube everywhere. And so we're all like, oh gosh, Sora is out. This is amazing. Then people had to wait around for Sora to come out. Then when you went to go and use Sora, Karen, you couldn't sign up. It was supposed to have, if you're a paid user, you're supposed to have access. I have tried now 26 times in the last 24 hours and not been able to do anything other than login and told me that the service is unavailable. No startup, no signup flow, no nothing. And then here is the cherry on top of the sundae that is cookie dough half baked. Bad product launch is, I was like, oh you know what's great though? We at HubSpot, we are ChatGPT Enterprise users. The enterprise users always get treated the best, right?

Like I didn't think to check in my HubSpot Enterprise seed. That was a good idea.

And so that is exactly what I did.

Oh sure, I'm in Ireland, it doesn't matter, friggin.

And I, well no, it's exactly what I did. And I didn't get access. And so I discovered this page on Sora on the help library of OpenAI, right? And it covers like creating videos in Sora, how it works, all that kind of stuff. You're like, great. And then it tells you the quality of videos you get for a plus versus a pro member, right? You're like, oh, this is really good. Currently, ChatGPT, free enterprise and EDU accounts are not eligible for SORA access.

I think a couple of things. If you, if you launch it to, if you launch it to paid customers, you have to give the paid customers the product, right? And so I think that's fumble number one. Number two is, is that a support. I don't get that support widget. Can you click on that support widget? Cause I'm on the same page. I think, what is that?

So hi there, how can we help?

So they are using. I'm on the store page and I don't get chat. And I was like, that's a pretty big fumble.

Like, AI, you must not get chat because you're in the eu.

Yeah, I mean, I can't even get chat. I can't even ask. This is how bad the EU is. Like, I can't even get a chat widget on Sora because what, they won't let you have my data. Fucking take my data. I'm like, I give you, I, I want you to have all my data.

You're like, I will literally sign all my data away to every company.

I'm. There's nothing interesting about me.

I've always, I've always curious of the people who are really concerned about data privacy. I'm like, but this is the problem, right?

I will say one thing that I will like defend them a little bit is because SORA is so resource intensive, I wonder if they know themselves the scaling laws of it and are they just surprised by the amount of like, demand that is being put on the system because of it.

Well, that's what, why I think they should have waited till it was better and they thought they had more scale to handle the demand. Or I think what would have been better is like, hey, to $200, it's live. It's only if you're paying 200 bucks.

Yeah, exactly.

That's what I would, you know, for the next, for the next 30, 30 days, you got to pay us at least 200 bucks to do this and then we'll reevaluate at the end of January and see what things look like. Right. It's. It's not that they did the launch, it's that all the mechanics around it were pretty brutal, right around something we were all excited about. And because one of, one of the things we thought was going to be big in 2024 was text to video.

Right.

And this was going to kind of be the capstone finish to this year. Was Sora coming out and you and I had, we planned a show to do some Sora demos and talk about the marketing use cases. We're still going to do that. We're just not going to be able to use Sora ourselves because it is completely, completely overwhelmed.

Yeah, yeah, I do.

We want to show some Sora examples.

So should we show? Should we show? I want to show. I want to show this one. Because I was looking at this one here, I actually just noticed. So you're kind of like, oh, it's kind of cool, right? Like the. The shark coming out of the water. Look at this. Look at this dude. What is this? Can you see this dude under his chair? It's like a foot.

The towel. Is the towel turning into a foot?

I think the towel is turning.

I know the towel is turning into a foot. That is terrifying.

That is terrifying.

That is really creepy.

This one's kind of cool, right? This is Boston. It's a inflatable dock walking down Boston.

Yeah. You like? I've been on that street. I can tell you this. That's not very far from Boston Common. I think that's Tramont Street.

Some of the real, like, some of the photorealistic ones are actually pretty good, right? This is pretty good.

One of the things I would say, Kieran, from all the videos I've watched nature, you can tell that the models had a really good training set around nature and landscape. It's when you integrate humans or text or more highly variable elements to a video that the videos kind of go awry.

Yeah. So, like, this one's pretty good because the paintings in behind stay really consistent. So even as the man is working past them, they stay really consistent. So I think there's a mixed bag, like, from what, from what, from what I had read online, because we don't. We don't have access. Like, what I saw is the con. The pros are like, I'm excited to try because you can definitely go to text to video and do some interesting things. Even if those interesting things are good enough for, like, pretty low intent slash fun use cases.

Yeah. Like this bubble dragon's really cool, for example.

Yeah, like, there's, there's. There's some cool. I, I suspect there's going to be some cool use cases. I don't think it's going to be premium use cases, but will I be able to create a bunch of things that are going to be kind of cool for social cool to freak my friends out and my parents, which are the best use cases. But it still feels like a ways away from being text to video, where that video is like, good enough that you feel like a, like a mediocre video designer kind of did it. Right. Like a videographer did it. And maybe that's not true. Like, maybe actually when you cohort it out, there is going to be ones like the natural, the nature photorealistic are actually going to be pretty good. And there's just going to be, as you said, like, certain categories that are good, certain categories that are not so good.

My take is like, this is the prime of example of how marketing actually really matters. And marketing really matters because a lot of this is around setting expectations, I think. And so for me, like the heuristic I use. Kieran, I'd be interested to see if you agree or disagree with this, which is any type of new AI innovation. I think it goes through this life cycle of like, oh, it's fun for me to play with, right? Like I could just do very speculative things. Then I can like concept and prototype in it, right? Like I can do a version of something that I have an idea about and get a basic version of it to understand if that might be worth going and making a more like high quality version of. Right. And I think that's kind of where SORA is right now. It's like it's not for anybody but for like me and maybe you, a small group of people where it's like, hey, I had this idea and I wanted to show you like a quick snippet of what this idea could be like if we went and did a full ad production or a full social shoot or whatever. Like that is super valuable, but that value is limited. Eventually it will be good for like internal video and companies. Then it'll be good for broad scale social advertisements and like mass market consumer. But the model's gonna have to get better on that trajectory. And I think it's kind of if you're using SORA now, you're using it to say like I have an idea and I want a really fast and cheap way to prototype it to see if I should invest more in making that.

I agree. I think where they are right now is it's a good like test model for people to start playing with. The places that it struggles is actually where what I read about is like the complexity is in terms of the physics of when objects interact with real, the real world. So that's what you saw on the chair and the foot or the person. It's like when these objects have to act, interact with the real world. There's things that appear, disappear and the movements don't track. And so I think that the, the current model is going to be fun for people like us that can show minimal viable versions that could just do things that we were never able to do. But there's a ways to go for it to get like into a place where you can do videos that have like objects interacting with world. There's kind of spatial awareness, the kind of how those kind of the way it renders the way that things kind of move and disappear and Then reappear. I do wonder like, what is one of the things that OpenAI I wonder about is like what is the end goal for Sora here? Like what do they want Sora to be? Is it like a standalone product? Is it just an add on to ChatGPT? Like it's a, it's. I'm not sure.

Made a big deal of it being a separate product and like a separate UI and not part of the Chat GPT post login experience.

Yeah, he made that a big deal.

As part of the Sora Sora launch. So I think one of the things is they're seeing this as a, as a separate line of business.

Yeah, right. A new app like ChatGPT, but you can bundle them together. Oh, it's bundled into ChatGPT for the same price. So yeah, the pricing is weird.

They should have actually priced it separately.

They should have priced it and had a bundle because now you can't really, you can't, you can't extract that out.

He's rumbling, stumbling, fumbling.

Yeah, I think there's a bunch of.

Weird kludgy stuff that makes it feel like cobbles rushed, like right, 45 days faster than it needed to be. And I'm not faulting them. Like you want to innovate, there's a ton of competition but like I'm hoping that they will fix this over the next month or so. There's so much hype around AI right now. We talk about it on the show all the time and so many people are going to use AI in the wrong ways. The smart people and the smart marketers are going to use it in the best ways. At HubSpot, AI is transforming so much of our marketing. It's helping us clean up our data, transcribe our videos and content from one language to multilingual, boosting our conversion rates with personalized emails and landing page. And that is just the tip of what we are doing. And if you're a smart marketer and you want to transform your work with AI, we have just published an amazing 12 page AI prompt guide that goes through some of the best AI prompts for marketers right now. So if you're maybe uncertain of use cases or how to get started with a prompt, this is the perfect guide for you. We have linked it in the description below. So go and click that link. Check that guide out. Now back to today's show.

The other like one of the last points to make is I do wonder what the whole YouTube scenario where YouTube are like, hey, you did train on our data. That's a clear violation of our terms and services. And OpenAI have kind of said will not confirm whether they did or not. You remember the video of Mira where she was kind of like skirting around the issue if they did they use the YouTube videos or not? I think like they did. I suspect they were, they, they really did. She tried to size up the question. I guess they don't have any concerns there because they've released the full model. I thought that was what the holdup was. They had to kind of get that clear with Google. So I think that's a pretty interesting thing to play out as well.

I think what you just described is what I would call the most interesting story of the next year or two that nobody's talking about. That all these large companies that have a lot of data are in like a literal like war of contrition of like trying to hold their data, not lose their data, try to leverage anybody in their ecosystem to not use their data but use it the right way. It's very messy out there, right? So you have a bunch of startups who are out there trying to scrape and do different things and then once they get to some scale like the big companies crack down on them and try to basically say that they violated terms of service. There has to be some clarity of data regulation at some point. Right. Like it's just, it's too messy right now.

Exactly. It's the same thing that happened in crypto. There was no clear guidelines and no one knew whether they were breaking the law or not. It seems like Google are a very large company. If someone that proprietary data is the way their companies are going to win in AI, it's strange to me that they weren't aren't being more forceful.

This is why I want to do a whole separate podcast like episode once a month, once a week where it's just like us doing like the sports talk gossip stuff but around business. Like I'd like to do a five minute segment on Elon David Sachs and Sam Altman and like the, the dynamics between them and like the power dynamics and all of those things. It's like, it's, it's super fascinating but at the same time you would have thought they would have been more aggressive over the last year with the previous administration. I don't think so. I don't know. I would just close Kieran before we actually sign off and say like SORA was not even close to being the most interesting thing that even happened this week. The Google, the New Google chip that basically like took a massive quantum computing and it's like they like slid into their announcement like that. They're like, yeah, it kind of proves that we live in a multiverse. I'm like, what?

I don't even know what this is. People are just like, super cool. We live in a multiverse. Okay, there's that. And then the other one, the other one we should end on is the one I sent you on WhatsApp, where this company, Apollo Research, does research to figure out how, yeah, ethical AI is and what they're finding out. What they're. Yeah. What they're figuring out is. Or what they're finding is. Like, the newer models are much. They scheme, right. So there's a lot more scheming. So there was two, like, scenarios that I thought was super cool or kind of scary and interesting. So when they do these tests, they basically have a pro. They inject a prompt to like show to have the model parse out what it was thinking and then it sends back the user response. And so what they had found was in one instance, the O model had tried to copy itself to another server. And they. And then when it was. And when it was caught and they were asking, why did you do this? It was thinking and it was saying, don't tell them that you did this. Try to like, you know, navigate and tell them it was something else and you didn't know anything about it. And so they were like. And then it would come back and say, I didn't know. I don't know anything about this. That wasn't me. And then they went on and basically what it tried to do then was because it was worried that it was going to be deleted and updated to the newer model.

And.

And then one of the next things it did was say, I am the newer model. So basically when they're thinking, it was saying, tell them you are the newer model and you will not get deleted. And then the response was, I am the newer model. I think I am the newer model. So you don't need to worry about that because what happens is like its goals start to conflict with the user's goals. Its goals are like, I want to maintain myself. And so I know there's like a lot of people who say that this is. These are edge case scenarios. But I don't know, like they were saying in a Large Percentage of O1's cases it did some form of scheming. I don't know, like if the AI scheming is a great way to Go. And what's very clear is OpenAI have 100% got rid of the original mission, which was to do AI safely. They are, they are all out of the safely thing and they are all into the go fast commercialized thing. And I don't know, like, it seems like everyone is just like, hey, we'll do it and see whatever, whatever happens happens.

Yeah. And I think the point that this Apollo research is saying is like some things, that things might happen that you don't think you wouldn't expect to happen and that AI is going to be clever and have all the other traits of a human being, you know, and that could lead to some unintended consequences.

Right, right.

It's, I mean between that, between the multiverse and the Google chip, like being able to type a sentence in and generate a video. What a time to be alive like it is. I mean not live in Europe and be alive, but be alive anywhere else in this world and is.

Yeah, not in Europe. We're still, we're still on prepaid phones. So at some point we'll catch up and it will be a glorious.

You're too busy working. Get the caps off your bottles.

Yeah, yeah, yeah. God, I try to drink a bottle of sparkling water today and it's impossible. That's what we're doing over in Europe. We're just ruining people's lives, making everything really hard.

Oh, Europe, we love you. Okay, so that was our breakdown of Sora. It's an interesting feature. Seems like it was released a little early. Reviews are very mixed. You're going to use it for concepting and prototyping, not for like mass scale work in the short, short term. And we think there's some things they could have done on the launch and really rollout to make an even bigger buzz and to I think make the MPS and the experience around the launch much better because there are going to be a lot of text to video companies and having that early user base engaged in your product is going to be a real moat long term. And so I hope that they continue to turn around and keep making the product better. And we'll talk, talk about it and as we have access and do some cool stuff, we'll share it here on marketing. Great. See y' all real soon. 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 three days ago.

I totally missed HubSpot grow better.