Did OpenAI Just Change Everything? Week 1 Review:Bonus Episode #artificialintelligence #ai

Foreign.

To the Next Wave. I'm Matt Wolfe. I'm here with Nathan Lands. And today we've got a very special bonus episode for you Today. We're going to cover the 12 days of OpenAI Christmas. I mean, I don't know what we're calling it, but OpenAI and Sam Altman made an announcement about five business days ago from the time of this recording that every business day for 12 days they were going to make some sort of new announcement or release some sort of new feature. And we wanted to make this bonus episode to quickly give you the rundown of what all of the features have been so far. We'll probably do a follow up episode once more of these special announcements come out, but let's just dive right into it and waste no time. Day one of this OpenAI 12 Days of of Christmas AI Santa Claus Santa Claus brought to us oh one and chat GPT Pro. So basically the quick rundown of what this is is they put out a new model or a new version of ChatGPT which is now a $200 a month plan, right? You've got your $20 a month plus plan. Now they have ChatGPT Pro which is a $200 a month plan which supposedly gets you even earlier access to new features, plus the most recent state of the art models like 01. And I believe there's even one called 01 Pro which is even smarter than 01. So that was day one. So Nathan, what was your take on it?

Yeah, to clarify, you definitely don't have to pay for the Pro plan to get access to 01. So. So before this they had released 01 Preview, which was a preview of their first reasoning model 01. And it was, it was pretty good. It was really, really slow. The new Zero1 is smarter and it's a lot faster. I mean it's actually usable now you can ask it stuff and you get responses back.

So like the, the O1 model is basically this model that's better at math and logic and, and that sort of stuff. I personally haven't noticed a huge difference from like what the O1 preview was giving us and what the O1 Pro was giving us other than O1 Pro seems to think a little bit longer and be a little bit slower. But like the responses from what I've seen so far have been pretty close to the same kind of responses as O1 preview. But I'm also like, I don't know, I struggle with like what prompt should I even test with it next? So there's There's. I'm definitely not using it to the max.

Yeah. Yeah. I mean, I would say O1 is a lot faster. Maybe you've only tried 01 Pro. Like O1 itself is way faster than 01 Preview. Like O1 Preview was very slow. O1 will. It's almost as fast as like GPT4O or Sonnet or one of those models. So I was really impressed with that. I've tried it for coding. As far as I can tell, it's the best for coding. Like I've compared it to Sonnet, at least in my own personal test. It's better than Sonnet and it's better than any of other, you know, OpenAI's previous models. But it is slow. So it's like, it's kind of painful. I'm hoping that OpenAI is going to release some kind of like model selector where based on what you ask it, it knows which model to use. So I'm not stuck using a one Pro on every single question or going back and forth.

I have to imagine that's going to be what's coming next, right? Or not what's coming, but what's coming soon. Like I don't know if that's going to be something within the 12 days of releases, but you have to imagine that that's like something that they're going to release at some point in the near future. You ask it a question and you don't have to decide which model to pick, right? Like it knows. Okay, this question needs logic or math or coding or whatever. Let's send it to the O1 model. Okay, this one is a sort of simple to answer question. Let's send it to just the plain old GPT4 model, right? It already knows that if you ask for an image to send that prompt to Dolly, right. I'm sure at some point Sora will probably worked into ChatGPT as well. And you tell it to make you a video and it will know to send it to Sora. I imagine they gotta be doing that with like the actual models you're using as well, but we'll have to wait and see.

Yeah, but I think, I think it's actually probably a bigger deal than most people realize just because like this is their first like reasoning model. And so this is almost like the GPT one of this. This is the worst it's ever going to be and it's very good at coding and people who are like scientists on X and stuff, they're posting saying for science stuff, research, that this is Incredible. Like, it's a huge step forward. So I think for daily, you know, people who are just asking it like daily questions like AI can already, any of the models can already answer, like stuff you would ask on a daily basis.

And I did personally go and grab the, the upgraded Pro plan for $200. But it wasn't because I wanted access to 01. It's because I thought maybe if I'm on their highest plan, I, I'll get stuff even earlier than everybody else. And also because I knew Sora was coming and I wanted to make sure I was going to be able to get into whatever Sora plan they had. So that was the reason I got in, honestly, for my own use cases. Like, I've had a hard time figuring out what to use these special O1 models for.

Yeah, that makes sense. I mean, I think for regular people, any of the models work now, like just for asking basic questions, you're going to see the reasoning. But I think at some point though they're going to like, like I said, this is the, this is the first version of their reasoning model. So like at some point these are all going to get baked into some kind of model. Maybe behind the scenes there's different models, but like we said, there's going to be something where you're, you're just using AI. And I do think, you know, a lot of people were saying, like when you look at O1 Pro and you, when you test on like a really, really hard questions, whether it's coding or science or anything like that, it's, you know, you could easily argue that this is already AGI. And so it's kind of wild that like the goalposts keep moving because people are not calling it AGI, but definitely two years ago, three years ago, people probably would have called this AGI. I really believe they would have.

Yeah, maybe, maybe. I mean, again, like you mentioned, the goalposts keep moving. Nobody seems to agree on the definition of AGI. Yeah, you're probably right. Like if somebody looked at this 10 years ago, like this is just mind blowing, uncomprehendable, like crazy advancements. Right. But since we've sort of gotten it like incrementally dripped out to us over the last three years, we've almost sort of taken it all for granted a little bit.

Right, right, right.

Well, let's talk about what they talked about on day two. And so day two was OpenAI's reinforcement fine tuning research program, which of the five days so far, this was probably the one that most people Were like, yeah, stocking stuffer. Yeah, yeah. This one's more of like a stocking stuffer sort of thing. This one seems to be more for like businesses, right. To. To. To sort of fine tune it on their own data. And it's a research program too. You have to apply for it, they have to let you in. So it's not something that any of us here have actually like played with or tested ourselves, but, you know, is a. It was an announcement that they put out during their 12 days.

Yeah, I don't, I can't really say much about it since I haven't personally used it. I think they were trying to sell it too, saying that it's easier to train, to kind of train it on your own data versus before. Training a model on your own data was a really long and hard process and apparently they've made it easier, but I haven't been able to try it.

Yeah, yeah. I think it's like a simplified fine tuning process really designed for businesses and researchers to kind of tailor the model towards like their specific niche or the specific information they're trying to train into the model. But again, this was like, this was like, well, we need to have 12 days of announcements. This was probably one of their more like throwaway announcements. They're like, well, we released the new plan yesterday. We released the new version of Open 01 yesterday. That was a big announcement. We're going to have a big announcement on day three, so let's just kind of squeeze a smaller announcement in between.

I'm sure for some of their enterprise clients, they probably loved hearing this. So.

Yeah, that's true. That's true.

That's not me.

So we're not the target market of this announcement. But then day three, we. We got another pretty damn big announcement, which was the day they finally released sora, which I'm sure everybody listening to this podcast knows exactly what Sora is. It's OpenAI's video model that they announced nine months ago and over the last nine months have sort of teased out some videos and what it's capable of, but didn't get it in the hands of most people. And then on day three of this OpenAI event, Sam Claus gave us Sora.

So, like, I actually managed to get into Sora. Were you able to get in or were you.

I was not able to. I, I woke up in Japan. I checked X. You know, that's. I've been checking every day, like, what's the new thing? And like, I'll wake up and I'll Say, oh, there's a notification from OpenAI. Okay, what is it? And so, and I, and I was, oh, Sora is available. And so I was super excited. And just like you, I signed up for a one pro for 200amonth. Yeah. Partially because I knew Sora was coming. Like, I was. One of my predictions, like, okay, Sora is probably coming. I had it number two and it came out the set. It was the second big announcement. So I was like, I nailed that. But then I couldn't get in. And so. And I'm still not able, yeah, I'm not able to get in. I've actually even chatted with some people at OpenAI and there's like, oh, sorry to hear that. I think access is going to be coming soon, but I can't personally do anything about it. I'm like, okay, so yeah, I'm not in.

Yeah, well, at the moment, right now, today you're not missing much because it's down, it's not even working. I can't get it to load. So like, you know, the big. So there was a few issues with Sora, right? They, they let people in. I, I managed to get in like the moment it was available, right. I was on Sora.com and when they announced that Sora was going to be releasing today, I was refreshing the page on Sora.com because there's a little button that said coming soon. And I was literally sitting there, refresh, refresh, refresh, refresh. And eventually I refreshed and it said, sign up now. And I was just like, boom. Like the instant it was available, I signed up because I just happened to be on the page, refreshing, waiting for it to open up. You know, I guess one of the benefits of being out here on the west coast was that I was able to do that. But then once I got in and started using was really slow, right? Like, to generate a video, it was putting you in a queue. And then it would take, you know, a 10 minute wait in the queue. And then it would take another, you know, minute to two minutes to generate. And then the second video I tried to generate, it was like 15 minutes waiting in queue. And then the third video I tried to generate, it WAS like 20 minutes waiting in queue. And it got to the point where like Even on the $200 a month plan, I was waiting 45 minutes in queue before it even started generating the video. And then like maybe 25% of the videos I tried to generate said unexpected error. This generation failed. And I'm like, I just waited 40 minutes to get an error that said generation failed. So, you know, and I was personally fairly unimpressed with the first handful of videos that I generated. The. The. A couple of the first ones that I did, like, I tried to do a wolf howling at the moon. That's like one of the first videos I always try to generate with all these video makers. And it basically made a static image of a wolf howling at the moon with like a sort of, like, parallax background where the stars looked like they were moving, but nothing was animated on the wolf itself. And I was. It was just kind of not that great. But what I sort of learned later on was that SORA really likes detailed prompts. Like, the more detail and length you could give your prompt, the better SORA seems to do.

But that makes sense. I mean, they've tested so far with, like, people who make videos, right. So, like, they're probably their, like, beta users were all, like, doing super detailed, you know, like this kind of. I want this kind of shot with this kind of camera and this kind of lighting and everything. Right. So that's. I think that's the people who've tested it so far. I would note, though, that this is actually the Sora Turbo model. Right. So the thing that they showed in February, the Sora model, I would assume, probably was higher quality and took a lot longer time to create a video.

Yeah, I think it generated up to a minute. These ones only go up 20 seconds on the Pro plan. And if you're on the plus plan, I think it's only 10 seconds.

Yeah. And Sam did allude to that. He's basically saying, like, hey, you know, this is the first version, and this is the first version that we're, like, showing the people it's going to get way, way better. And I'm sure they, you know, a lot of the. I doubt they just worked on Turbo. I'm sure that they probably already have, like, Sora 2, you know, internally.

Well, have you probably even seen the video going around on X right now where somebody's, like, supposedly demoing Sora 2 already?

Oh, yeah, the Viking stuff. Yeah.

Yeah. I don't know if that's real or not. Like that. That could just be, like, fake. I have no idea. But supposedly people are already seeing Sora too.

Yeah. And that looked incredible. Like, they had the. The Viking woman with, like, the blood all over her face. And it actually looks close up on her face with her. Her eyes, everything. And it was like. Wasn't perfect, but, God, it was like 99. It was like, that could have been a shot in a film.

Yeah. I think one of the bigger issues with Sora right now, you know, we were talking about this before we hit record, was they announced Sora nine months ago. And then in the last nine months, we've gotten new video models from Luma, we've gotten new video models from Runway, from Pika, from Cling, from the. What is it? Hello, AI. I don't even know how to pronounce that one.

Yeah, there's a lot.

The one called Minimax, and you've got one called, like, Hot Shot AI. And there's all of these video models that have come out since we've seen Sora that sort of almost got or even surpassed what we saw from Sora. And so by the time Sora actually came out, I went and put a video up on my YouTube channel about it, and it didn't do really great. And the comments weren't that great about, like, what they saw from Sora. And I think a lot of that was just like, well, you gave us, like, nine months to marinate on it. And you gave us. And you gave competitors nine months to make other AI video tools. So by the time you finally drop Sora, people are like, cool, glad you finally. But we're past it.

I think Sora will be the first one that's actually used, though, like, for, like, commercial uses. I really think so. I saw some comparisons of Sora and Runway, and Runway has some great editing tools, and Sora has some great editing tools. I think they're both, like, the top in terms of, like, the editing tools now. But Sora's shots looked better, like, especially with, like, people. Like, there was. I saw some shots comparing, like, okay, create me a scene in a cafe. And the shots by Sora looked way better than Runway to me. And the Chinese models, I mean, you gotta realize, like, they're building them in a. There's a lot less restrictions for them, what model, like, what data they're training on. But also any kind of safety stuff, right? Like, exactly. They're not having to worry about that stuff. So I'm pretty sure that, you know, there was already leaks that open. AI was talking to people in Hollywood, I think, back in, like, I don't know, what was it, like, April or something like that. So I assumed, like, the first. You know, the first people, like, definitely, Hollywood's gonna be using Sora before anything else is what I believe.

Yeah, I mean, we've already seen it. Sora was used in a Toys r US commercial. I think that Coca Cola commercial might have used sora. I think we've already been seeing it used like in professional productions a little bit.

Yeah. But definitely they botched the launch like big time. Like it's super delayed and then it finally comes out. And I don't think people also really got that this is the Turbo model, so it's probably lower quality than what was previewed in February, you know, and then like.

But it does have cool features. It's got like a blend mode where you can take two videos that merge into each other. And yeah, you can remix existing videos.

And yeah, remix has been getting popular. Right. People are showing. I've been seeing that even like with like the best Chinese models, you can produce videos with them and then put them into Sora and do the video to video and then the outputs better than what the Chinese bottle gave you.

Yeah, I mean, I think Sora is, is good. I think I've gotten some bad outputs from it, but I think it was more my prompt than it was actual Sora. I think the more detail you put into it, the better it's going to be. But, you know, it's the worst it's ever going to be. Right? It's. It'll. It's only up from here. It's only going to get better from here. I think the full Sora model is going to generate one minute videos at much higher quality and hopefully they get their servers under control and figure out how to make it so more people can use it.

But also, I mean, like, yeah, they really messed up the launch, but also that's probably a sign that they're going to win in terms of distribution. Right. Like, they probably have a lot of people trying to use Sora I haven't heard of. You know, like, yeah, sure, cling and you know, Runway and Pico, they all had minor issues when they've launched, but they didn't have these kind of issues. And OpenAI has a lot more funding behind them and access to servers, so they're probably having a lot of people trying to use it.

So let's go ahead and move on to day four here. And day four was when they announced Canvas. But we've already had Canvas. If you're one of the paying members, if you're one of the pro members or plus members, Canvas was that feature where you can have it generate code and it generated code in its own box. Now Canvas can actually run the code. That's one of the new features of the new version of Canvas. Is it can actually run Python directly in your browser. And that wasn't available before. So this is basically OpenAI and Chat GPT catching up with Claude, because Claude's had that inside of Claude artifacts for a little while now, right? You have that little separate sidebar, you can see the code. You can actually run the code directly inside of the window.

And yeah, it's a little different.

We have it in ChatGPT.

Yeah, it's similar, but it's a little bit different. So, like, it's not, it's not like with Claude, where you can actually see the output of like a website or something like that, or like an app you created. It doesn't do that. It'll. If you do like a basic Python script that has some basic output, like text output or something, it'll run that and you can test the code. And actually it's something pretty cool they're doing. They have like syntax highlighting as well. So you can actually see like different colors for different parts of the code. Like, here's a variable, or, you know, whatever. Here's an if statement. It's doing stuff like that as well, which is really cool. But also this is available in custom GPTs, which I thought was interesting. You know, before it was not available in custom GPTs, and I tried. I have this one custom GPT, it's like a social media assistant or something like that that helps me. Basically I created it to help me create tweets and help me create LinkedIn posts, and it's way better now with a canvas. So when I, when I edit a tweet or edit a LinkedIn post now, usually, like, it would give me something, I'm like, oh, okay, parts of this I like, but parts of it I really hate. And so I, I'm just gonna change it myself. But now with the canvas, you can actually go in there and like, you'll, you can edit it together, basically. So it'll be like, here's the post. Like, you know, maybe, maybe I'll give it like a news, a news story and like copy and paste it in and say, hey, help me make a LinkedIn post about this news story. And maybe it gives me some parts I don't like, some parts I like. And now I can kind of type in there and kind of interact with the. With ChatGPT while I'm editing it, which is really cool. Like, I found that super useful. So to me this shows that long term, you know, it kind of makes me worry for like startups like Cursor, you know, like the AI coding tool and things like that. It kind of feels like at some point Chat CBT is going to have their own interface for writing and coding.

And I think the other thing with Canvas is they rolled out this Canvas feature into the free plan too. So if you just use Chat GPT free version, you can now actually use Canvas. So it was only in the paid versions before, now it's also in the free versions.

So it's not available for Pro, which is annoying. Like, right now, Pro is like the smartest, best model. But then if you actually want to write something and like interact with it, which is probably the best experience, is now to go back to 4.0canvas, which is kind of.

Well, you can still use it if you're a Pro subscriber. You just can't use it with the 01 Pro model.

Right? Yeah, correct. Yes, correct. Yeah.

Cool. Just wanted to.

Yeah, it's super annoying, though, because actually O1 Pro is actually pretty good at writing. Like, I've tested it. It's pretty good at writing, especially if you. If you. If you get references, if you give it, like, style references, it's way better. And so I think possibly if you give it style references, it's the best at writing right now. But then you can't, you know, but then this new feature, Canvas, is super awesome for writing and you can't use it with the One Pro.

So moving on to day five, which is actually the day that we're recording this, they announced that Chat GPT is now available on iPhones. This, to me, was like, they knew that iOS 18.2 was going to be released today by Apple, so they use that as, you know, their Day five announcement. And they just line it up with the day that Apple was going to announce 18.2. Right. Because on the day we're recording this, Apple 18.2 came out, which has all the chat GPT features in it that people can now use Siri to call upon Chat GPT. But other than that, Chat GPT is also available as like a Mac version of Siri. So if you use a Mac computer, you can use Siri on your Mac and invoke ChatGPT to, you know, run things and to see your screen and things like that directly from, like, the Siri version on your Apple devices. So that was the big announcement today. It was more of like a. An Apple announcement that Chat GPT kind of borrowed the announcement of.

Right, right. Yeah. I mean, it could make ChatGPT go way more mainstream, though, in theory. I'm not sure. I Haven't had been able to try it yet. It seems kind of weird. So, like, what, you talk to Siri and then if Siri can't handle it, it's like, hey, let me go get ChatGPT. It's kind of, it's kind of a weird interaction, if that's how it works.

Yeah, well, in the video they showed this morning, they showed it where you can basically turn on or off the ability for it to like, confirm that you want to use ChatGPT. So if you turn off the ability, Siri will just decide on its own whether to respond using Siri or, or whether to respond using ChatGPT. Or you can toggle a lever and if you toggle the lever, it will ask you if it's okay to go ask ChatGPT. So you kind of have the option to turn that on or off. But yeah, it's like it's going to try to use its own Siri model first and then if it can't, it's going to go to Chat GPT. Or you can specifically instruct it to ask Chat GPT. So you can go like, you know, hey, Siri, ask Chat GPT this for me. And it will go ask ChatGPT. Now, my series going off on my desk here. But that's, that's essentially what it'll do for you, which is cool. But, you know, the new Siri features, you have to have an iPhone 16 or newer. I don't have an iPhone 16 or newer, so I haven't been able to test it yet. And I'm not on a Mac computer, I'm on a PC, so I haven't been able to test the desktop version yet either.

I'll have to test it out and see for a comment on, I guess.

But Those are the five days of OpenAI announcements that have come so far. We still have seven more days. I'm sure we'll do like another bonus episode where we break down the other seven days. But since you mentioned it, I'm curious, why don't you break down your other predictions of what you think OpenAI is going to launch during the next seven days and then when we revisit this next bonus episode, we'll see how close you got.

Yeah, well, I mean, I think eventually, probably in the next year, they're going to come out with a new model that bakes in. They're going to combine the best of a normal LM with a reasoning model, and that's all going to be combined. But there is the potential that they still have like a GPT 5.0 coming out. That's just like a basic, fast model that's way better than, you know, because people are still, like, people who are in the know are still saying that Claude, you know, Sonnet is the best, like, for just like a really fast model to get basic questions answered or even for coding. So I think it's, you know, it's still possible, the Orion model, but, you know, I think. And so a lot of people are kind of expecting that. I think what's going to be more interesting is if they have like an actual really useful version of computer use where the AI can actually use your computer for you and do things. I think, I think that's probably what they're going to release. I mean, I assume, like, I assumed that Sora was going to be one of the first, first two things they released because I think they have a lot more big things coming. Because, like, Sora is like, okay, yeah, like, it's exciting, but people already knew about it, so it's not, you know, I don't. I don't think they would let that be their big grand finale. So I think the grand finale is going to be either computer use or agents or maybe both. They're like, they really kind of combine, right? Agents that actually can go do stuff on your computer or do stuff on the Internet for you. And, you know, it was rumored probably like five months ago that people had seen demos from OpenAI in San Francisco of them having agents that actually could go off and do, like, work, basic work for you. So I would not be surprised if they have something that's gonna be really impressive with agents that, you know, the kind of thing that companies could, you know, actually start using on a daily basis. Maybe something with email, like it handles emails for you. That'd be obvious, like a basic thing.

The only thing that I'm fairly certain that we're gonna see over the next seven days is I think we're going to get vision inside of the advanced voice mode. Because if you remember when, when they showed off advanced voice mode, a lot of what they were doing was like taking pictures of math problems and asking it to help solve the math problems and, you know, basically pointing their camera at things. Kind of like what we've seen with Project Astro recently, pointing the camera at things and then talking about what the camera seeing. They show that off when they showed off advanced voice mode the first time, but we haven't gotten that in advanced voice mode. Other than that, I don't know what else.

Yeah, yeah, that's a given. It's like Greg Brockman, he was on God. Who was he interviewed by? But he was interviewed recently and he showed a demo of the computer vision stuff. So I'd be shocked if they don't update advanced voice mode with vision and hopefully they improve it. Cause like, you know, now that I've gotten my hands on it, like, I love it and I hate it. Like, it's like it's so good in some ways and then it's still, you know, it's like it's. If there's silence, it wants to keep talking to you. And it's like, God, let me think for a second. You know, it's like it's really bad about that kind of stuff. It kind of sucks at translations actually. Like, it's good in some ways, but then it gets confused going back and forth between the languages. But it's like so close to being really amazing and really useful for me, but it's like it's not right there. So I'm hoping they improve it as well. Not just the vision, but some kind of update to the model, you know.

That's our recap of the first five days of OpenAI's 12 days of announcements. Again, we will do more follow ups and talk about more of the other announcements that come out, but we wanted to quickly make this bonus episode and run down what, what we've seen so far and what our thoughts and what we think the implications are. And hopefully you learned a little bit more about what OpenAI has been up to and where they're going with all of this. If you enjoy episodes like this, make sure you subscribe to us wherever you subscribe to podcasts, preferably YouTube, we're doing a lot more visual stuff on these podcasts, but we are available wherever you listen to your audio podcast as well. And thank you so much for tuning in.