Ep. 13: What is Sora?

[00:00:00] Foreign. [00:00:05] Welcome to ChatGPT Curious, a podcast for people who are, well, curious about ChatGPT. I'm, um, your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what ChatGPT actually is really though, are the files in the computer, how to use it, and what it might mean for how we think, work, create and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it. [00:00:38] Hello, hello, hello, my curious people, and welcome to episode 13 of Chat GPT Curious. I'm, um, your grateful host, the Maestro, and today we are talking about Sora 2. [00:00:49] So whether you like AI generated videos or not, they are here in a very big way and they are here to stay. So. So I figured I'd use today's episode to take a dive into OpenAI's major, major, major update and, uh, their newest rollout, Sora 2 and its accompanying app, Sora. [00:01:11] So for those of you who don't know, Sora 2 is a text to video diffusion model. Big words. We'll get into that. Uh, but you give it a prompt, it generates a fully animated, fully photorealistic video clips, including sound like, this is from the future. [00:01:30] So I want to nerd out for a second before we kind of talk about the app and, and, and that stuff. And I want to talk about what, what this thing is like, how it actually works, right? Because we are curious humans. So when I was reading up about sort too, I saw this word diffusion, right, is a text to video diffusion model. And you'll see that word diffusion everywhere if you go and search about it. And I was like, what the fuck is that? Right? What does the fusion mean? So I'm going to attempt to explain it without getting too in the weeds. But I do think that when I'm sitting here and outlining this and reading about it, I was like, this is pretty cool that we. I'm going to use we meaning you and. And I, right? But you folks that have, if, uh, you've been listening to the episodes and maybe you have to listen to them more than once, but if you've been listening and you're like, yeah, you know what, I can, I can hang, right? We can understand this stuff at what I would consider to be a somewhat intermediate level. It's not like a cursory, superficial level. Like we, we can understand this. Uh, and no, we don't understand necessarily the math behind it, right? There is. And I have to give props or props too. Like that's one of the things that excites me about, about AI in general is that the math and the intelligence, the brilliance behind this is incredible, which gives me hope for humanity because I'm like, fuck, we're so smart. Like people are so smart, right? But no, we don't like, know the you and I, maybe some of you listening, but I don't know, definitely not me don't understand, like the math behind it. I read these articles, an article from IBM and it was talking about it and I was like, wow, this is the math doesn't have. The math has letters in it at this point, folks, like, it's all letters. And I'm like, wow, this is, this is a lot. I know you folks remember if you went through higher level math, you know, gets in, you get into college and like that's the last math. I remember taking that, I was like, where are all the numbers? Like, where did the numbers go? And then it was, it was too much for me. But I do think it's incredible and it does give me hope for humanity because we are so smart. Like humans are so, so, so smart. But overall, no, I am not stoked about our ability to make these AI generated videos. [00:03:27] Uh, but pretending that, you know, does. The technology doesn't exist. It's not gonna make it go away. So let's at least understand what's going on. So diffusion, let's talk about that, right? The SORA 2 is a text to video diffusion model. [00:03:41] What does that mean? It is called diffusion because the core idea, the central idea is modeled, it's borrowed from physics, right? It's borrowed and modeled on physics. Diffusion process. The way particles heat, right? Particles or heat or ink, whatever you want. How it spreads out over time, right? Things naturally move. Remember going back to the science class, Diffusion, things being moving from areas of high concentration to low concentration. So think about a drop of, of dye and you drop it in water and it spreads out, right? It's a fusion. [00:04:11] So that is part of how this model works, right? These models, during the training phase for these models, that's literally what's happening. It takes a video, it's fed a video and it diffuses that video, right? It creates randomness, disorder, chaos if you will. It diffuses it by adding noise. Now again, this is all math. So if you all just want to be like, yo, it's math that's happening, that's, that's fine, right? But if you think about like, if you're trying to like visualize this in your head because it's how my brain works. If you think about taking an image like uh, that's on tv, diffusing it would be gradually turning it into like static or snow, right? Going from a super sharp image and going. And it's like, oh God, what's happening? And it's turning into, you know, snow. The white stuff that's on the tv, that is diffusing it, right? We're adding noise, right? Of, uh, note, just to keep accuracy here. For those of you that are nerding out, the image itself is not actually being diffused, right? But rather the data and the math that creates those images. But if I was to say for if you're like, fuck this, my head already hurts, just think about taking a sharp image on tv, it's crystal clear image and turning it into static, okay? That's diffusion. [00:05:16] That is the first part of training, uh, the first part of training is taking clear images and turning them into snow, right? Adding noise. And then the model learns how to reverse that and create a clear image from the snow that is there, right? It's removing the noise. [00:05:32] Being able to generate a clear image from snow, that is called denoising, right? We're removing the noise. And that is actually how videos are generated. Which is actually really cool when you think about it right now. Why does this have to happen? Because video generation doesn't start from a blank canvas. Like we kind of think about it like, oh, there's nothing there. And then like there's a video, but it actually does not work that way. Right? Think of it more as uh, starting with a canvas that has everything on it, but it's all fucking scattered and random and chaotic. And the goal is to organize the chaos, AKA Denoise it and create the video that you want, right? Based on the prompt that you input, right? So the perfect analogy. And I say it's perfect. Cuz I asked Chat GPT and I was like, can I think of it like this? And it was like it literally gassed me up. It was like, this is brilliant. And I was like, thank you. Right? So the analogy that I, that I think of in my head and if it works for you, amazing. If it doesn't, I'm sorry. Um, but the scene in Willy Wonka, right, where the kid, Mike, right, the cowboy kid, and he has the guns, right? He gets turned into that like colorful like cloud of, you know, dots, like the cloud of colorful dots in the, in the sky, Right, Right. It's like I Don't say the sky in the inside of the room. And then he gets transported into the TV. [00:06:42] So image generation that Sora does or that Sora 2 does is like starting out with all those little dots, all those little particles, and then generating a mic, right? Uh, the training phase would be the first half of what happened where Mike gets exploded into those little particles. And the model, it studies what Mike looked like, right? What all those particles when they were, you know, in order, what it looked like. And then it learns the logic of how the dots used to fit together, right? That is the training phase. And then the image generation phase is. [00:07:14] Imagine all the little dots in the sky and the dots in the. You know, they're in the top of the room. And look above their heads, right? Uh, if you've watched in the movie, you know what I'm talking about, right? They're all above this head. And then it turns into Mike into in the tv. That is what's happening, right? We are not starting with a blank canvas, right? We are starting with a canvas that has everything on it, but it's all scattered and random and chaotic. And the goal is to organize the chaos, AKA Denoise it, and create the video, right? Based on the prompt that you gave it. It is all math. It is really amazing, honestly. [00:07:46] But it's also a disaster and you probably should not have this technology. So, uh, that is how Sora 2 and any diffusion model works, right? [00:07:56] Training, turning the images into snow and learning how to turn them back into clear images. Generation is starting with snow and creating a clear image. [00:08:05] Now that we have taken the deep dive, the pseudo deep dive into how Sora 2 works and how any diffusion model works, let's jump into what SORA2 actually is and what Sora the app is. Right? So simply stated, Sora2 is OpenAI's text to video generator. You type a prompt and it creates a Sora super realistic video clip with sound like talking, whatever you want. It's. It's actually crazy. As of right now, the clips are limited to like 10 seconds, but this will likely change by the time this episode airs. So, like, don't hold me to that. But as of right now, it's 10 seconds. Uh, Sora 2 accepts both text and image inputs and it can handle. This is new. It can handle camera motion cues and environmental detail controls. Like, you can really prompt the out of this. So I know I've talked in the past about folks wanting to do the most with prompting and how like it, it's, you know, becoming largely Unnecessary to, like, be so crazy about it. But this is a time where it's actually really helpful. Like, you can be like, I want the point of view to be this. I want the camera to pan in this. Like, it's actually pretty amazing, right? So notice it is called Sora 2, which means that there was a Sora 1, which maybe you didn't even know existed or notice existed. Because if you've been using Chat GPT, it's been right there. The original Sora was inside of Chat gbt. It was on. It was in the menus on the left side. And honestly, I never used it. I. I'm not a fan of AI generated video. I have no need for it, no use for it. Uh, but it's been there. So Sora as part of ChatGPT, was first available in December of 2024, uh, which is, like, not that long ago, right? Less than a year ago. And it made the type of AI videos that you'd expect, right? Where you could 100% tell, like, that is AI. Like, what happened. This is just like, this is nonsense. Sora 2 rolled out on October 1st of 2025. So just, you know, from the day that this dropped, I don't know, like two weeks ago, they removed Sora from inside of Chat tpt. Like, they unbundled it and they gave it its own app called Sora. Why is it called Sora? I actually couldn't find out why. [00:10:06] Maybe, possibly what I found. What I found is that maybe, possibly it's because Sora means sky in Japanese, and that speaks to, like, the bigger picture and wide open spaces. I don't know. Uh, but I couldn't find anything concrete regarding naming. But it's called sora. Okay, so 2025, October 20th. Wow, wow, wow. October 1st, 2025, rolled out. This thing is brand new and the, the. [00:10:28] The advances are just kind of insane, right? So the app itself, Sora, the app, as of right now, it is invite only. And what that means is that someone who has access to the app has to invite you by sending a code. It's smart that they're doing it this way, right? It creates demand and, like, people like, I want it, I want it. Um, so Forest, my guy, Forest, actually invited me. So I, um, I also have three remaining invites. So when you get invited to the app and you sign up, you get four invites. Uh, I sent one to Lex. I don't know if she used it yet. Um, but I have three left. And if you want one DM me or text me 310-737-2345 DM at the Movement Maestro first uh, come, first serve. Okay, um, but the interface looks basically like TikTok, right? The interface for Sora. And some folks have lovingly named it Slop Talk, right? Because it's just like, you know, some of these are just, it's just you. I don't have to even use words. I don't think you folks know what I'm saying. Like, yes, it is hyper realistic, but it's also like, what the fuck are we doing? Right? [00:11:32] What the fuck are we doing? But looks like TikTok it is vertical videos. That's all it is. And they are all AI generated the captions of the videos. Uh, on some of them are the prompts that were used to generate the video, which is actually pretty cool because some of them are extremely long and like in depth. And you're like, wow, that's that. Yeah, that makes sense. Um, and it also most notably has a feature called Cameo, which is how you can insert yourself or anyone who has set up their cameo account into the video, right? You want to put yourself playing basketball with somebody or hiking Everest or doing some shit. That is how you do it. [00:12:10] The biggest issue, I know this bajillion issues, but most notable issue, at least for now, is that it has an opt out policy which is pretty whack. So instead of opting in and being like, I consent to having my used. [00:12:25] The default for Sora is that users must opt out of having their used. [00:12:31] Um, I couldn't even find out how you do this. Like I don't know who you write into or what happens. Like, I don't know. [00:12:38] Uh, you can opt out for the training, like training of the app the same way that you can do that for Chat GPT, right? Where there's like inside of the, um, there's a toggle that says improve the model for everyone. And so if you go into the data control section of the app, you can toggle that on or off. It was off on mine. I don't know if it's because it's like tied to ChatGPT or whatever, but you can toggle that off there and that is that you like don't want your used to train it but like for the actual content generation side and opting out, I. There is no clear way to do it and this is going to be a problem. Like it, it is, mark my words, it's gonna be a problem. So we shall see. Um, as for Sora being the, the app Being tied to your Chat GPT account. It is supposed to be, though, when you sign up for Sora, it does make you create a new Sora account. [00:13:27] Um, but then, like, I went into the account section and it like had my email and phone number and I was like, but I didn't put that in there. So how the fuck did you get that? Um, so I did. I was asking ChatGPT. I'm like, how did it know? And it said that it could be from the. The device id. [00:13:41] Nothing is private anymore, folks. Nothing. Nothing at all. Uh, so, you know, I'm not telling you to go and like, put all your everywhere. Like, I, I do this episode, I do this stuff to just to make you aware of what's going on. But I did try it. I did sign up. I do like to just in general take, uh, my username Movement Maestro just on all the platforms. Like, it's just something that I like to do. Um, but yes, that has been my experience thus far and either way, I do anticipate a lot of lawsuits coming down the pipe. [00:14:07] Um, but I will say that overall, having Sora as a standalone app is actually a very good thing in my opinion. Right. If we have to have it at all, I'd rather just not have it. But if we are going to be leaning into this, if this is going to be a thing, which it is, it is better that it is a standalone app. This way you diff can differentiate between AI content and real content. You don't have to think about it. Like, if you are on Sora and watching it, it is AI. That's all that's allowed to be on there, right? It's A.I. generated. [00:14:34] Um, the issue is that you can download these videos and then upload them to any platform you want. It does have a watermark on it, at least for now, that says Sora, but people don't know what that is. And so I'm actually seeing, because this thing just got launched, I'm seeing actually videos that people are responding to. My guy, my guy, he doesn't know me. Tony Baker, he's a comedian. He does like these animal voiceovers and he did a reaction video to this. This video is clearly AI generated. We've seen a lot of these going around though, right? Like, this is not super new, but it's like the dog that's living with wolves and you're like, that's clearly AI. But like your grandmother don't know that. And that goes viral on, on Facebook and then like, suddenly it's everywhere. So you can put these things, you can download it and put them on other platforms. There's a Sora watermark, at least on these. [00:15:21] But I'm wondering actually if it's not like in the middle. I mean if you could resize it and like you cut that out. But I mean it's all, it's a watermark that's on there for now. Like how hard is it to get a watermark off of things? Things like we, that's if it hasn't, if there's not an app for it yet, it will be here in the next five seconds. So, uh, I like that it has its own app and this way, you know, if you're on it, it's all AI but as soon as those videos go to another app, which they can very easily, then you're, you're kind of right, like that. [00:15:51] I, I, I'm like at a loss for, for words because I'm like, I want to report on this and tell you what's going on. And I'm also like, this is not very good. Right. Overall, I do not think that it's a good thing that we have this. Um, it's clearly super compute heavy. Like I don't have any numbers, but like you're generating a video. Right. Oftentimes the argument that's used for chatbots not using that much energy is to compare to other things, compare them to other things that we know. And uh, one of the things we compare them to is streaming services and how streaming services use so much more compute and just everything. Right. And now we're out here making videos. Yes, they're 10 seconds long, but that's just for now. Like it's definitely going to go up and you can just make them so easily. You're just like, it's super fucking simple. Like there's no coding. You just type in any prompt and it makes a thing in like two seconds. This is going to be a problem, right. Additionally, it blurs the lines between real and fake. Right. We have basically given everyone with a phone the ability to make deep fakes from their phone. Right there. There's super realistic videos on there of, of Dr. Martin Luther King Jr. And he's talking about video games and Michael Jackson and he's stealing chicken, just stealing fried chicken. Like you can make anything on there, anything. And you know the video is being super realistic. Even if someone proves that it's fake, once it's been seen, it's been seen, the damage has been done. [00:17:09] This is going to be a problem. I also think it's worth noting the jump in quality and how quickly it's been less than a year. [00:17:16] And this video, this is like, like just insane advances like. Yes, I know I have to play advocate there. Devil's advocate though is, you know, Chat GPT. I think I talked about this in, I don't know, one of the last few episodes. Just ChatGPT has, has definitely stalled in how quickly it's improving. [00:17:36] Um, but still, man, these videos went from being very clearly AI to being like, wait, I don't, I'm not sure. Especially, especially when the video doesn't have a person in it. Then it's really hard to tell. Really hard to tell. So to summarize, OpenAI rolled out a new and improved version of their text to video generator and it is called Sora 2. It lives inside of its own TikTok like app that is called Sora which uh, is currently invite only. Let me know if you want an invite. Uh, the model creates super realistic videos with, with sound like that's insane to me. And the model works via uh, diffusion. I foresee many, many, many, many, many lawsuits and uh, more bad things than good. But here we are. All right, so am I telling you to go out and use it even though I'm giving out codes? No, I'm not. Uh, I'm curious. You're curious? And I just want to make sure that you know what is going on in the world of AI. All right, last things last. Before I wrap it up, how I use Chat GPT this week. Each episode I include a section where I briefly discuss how I use Chat GPT that day or that week. So this week I'm actually gonna share two things number. Uh, one, I use Chat GPT to generate the prompt that I put into Sora. Yes, I made a video because I wanted to see what it can do. Um, but no, it's not me. I did not put my felt myself in that, my face on that. I had uh, it make a beach scene and I was like, oh, that's pretty cool. If you want to see that video, I put it inside of Vimeo so you can go check that out. It's literally 10 seconds long. But it was pretty wild man. Like to just be able to type in a uh, little prompt and it comes up. So if you want to check it out, there's a link in the show notes. Um, you click on that and it also the prompt that I used is in the description for the video. So that's one way that I use Chatgpt Second way is that I use it to help replace. Use it to help me replace the power board on our wine fridge that has no wine in it. Right. We don't put wine in that. Champagne problems, because that's all we put in there is Prosecco. Uh, but it started glitching. The one. It sounds so bougie to have it. I'm not trying to, like, justify that we have one, but I would never, of my own accord, have one. Like, we moved into that. We rent, and there was one here, and so we just use it. And you can't, like, get rid of it because it's, like. It's, like, built into the. It's like, slides into the cabinetry area. So, like, you don't have to use. I guess you could just turn it off. But, um, it's there. So we put the Prosecco and stuff in there because we bring the Prosecco. [00:20:01] We, uh, being me and Lex, uh, Lex and I bring the, uh, the Prosecco whenever we have birthdays and such to celebrate at volleyball. That's. That's. It's a tradition that I started. And we keep doing it where just bring, uh, mimosas. So we always got a bottle. Always got a few bottles of Prosecco on hand. But it started glitching. And, like, the front. The. The part that lights up where the numbers are was like, all, like. Like, freaking out. And so Lex got a repeat, tried pushing some buttons and unplugging it, the usual, and it didn't work. So Lex called and got a repair guy out with Send m me my box. He futzed around with it for, like, an hour and then was like, drilling stuff, and what the fuck are you drilling in there? [00:20:36] And. But then basically it was like, I can't fix it. It needs a new part, and it's gonna cost 500. Literally quoted me 500. And I was like, immediately, no, that is the price of a new one. Uh, so I asked him. I was like, what's wrong? And he's like, it's probably the power board. Like, these things can happen. And I was like, all right, cool. Like, I'll just get a new one. So I went online and was, like, looking at them, and then I was just like, I bet. Let me. How much is the new thing? Like, I would probably fix this thing. Like, if that man could fix it. Honestly, Come on, guys. If a guy could do. If a man could do it, how hard is it? I said what? I said, uh, but I was like, I think I could probably fix it. So I was like, lex, let me see if I can. What do you think? Let me see if I can fix this thing. If we can't, then we're just out the cost of a new power board, which is like 50 bucks. Like, like, let's. What do you think? And she was like, okay, sure. So I went and bought a multimeter. Um, I chatted with chat GPT first, and I was like, what could be wrong? The guy said, it's this. [00:21:29] What would it cost to try and fix it? Like, does it make sense to try and fix it? And ChatGPT knows that I'm like a DIYer because I talk to this all the time. And it was just like, honestly, it's worth it if you're quoting 500. Like, you need a multimeter. And then it found the power board. It's like, the powerboard is 50 bucks. Um, then it gave me a few options for multimeters. And I was like, well, I don't need a super expensive one. And it was like. I think it was like, the one that I ended up getting was 80 bucks, only because the one that was cheaper, they didn't have it at the local store near me. So I didn't want to drive all the way to Home Depot. There's a hardware store. It's like, close, like a local one. And they had the $80 one. I was like, fine, whatever. I'll just get out. I think it was 70 bucks. I needed one anyway. Uh, it's good to have one for little tasks. So I went and chatted with ChatGPT, got a multimeter. Um, we figured out what was likely wrong using the multimeter. Right. So it wasn't for sure that it was the power board. Um, and I wanted to confirm that. And so, so I needed the multimeter was to see, like, where is the problem? It's clearly an electrical issue. It's not the compressor. It wasn't like it was turning on and off. [00:22:22] Um, it was like. So it was like, is it getting power? Like, is there something on the actual board? And it. It's. The logic is. And so we figured that we chat GPT. And I was like, seems like this is the issue. [00:22:33] Uh, so I ordered a new power board, which it found for me. It was like, hey, as I put in the specs for the model and everything, and it was like, here you go. Uh, I ordered it. That part came in like two days. It's a local company, actually, I learned. And it's like, literally like some small company. And they sent It, I installed it. I, you know, had Chat GPT help me out. It was very simple to install. I just like unplug um, these wires and put these other ones in and that works, works brand, it works right as rain. [00:22:58] So you know, to me, one of the best ways that Chat GPT can be helpful is in diagnosing a problem, right? Because, and this is like as a pts, my background, if you don't know, like I'm a physical therapist and like that's, that was always the most fun part to me was assessing like what, what the. Is going on here. Because once you know what's wrong, then you know as it relates to these, these DIY things, then you can lean on things like YouTube and Reddit and, and you can ask ChatGPT to help to find the things on YouTube and Reddit to give you the, the how to fix it. Right? ChatGPT can also do it, but sometimes it like it's nice to be able to see a video of the thing, right? [00:23:30] So the diagnosing part you can use ChatGPT for. I will say though, like I've always said you got to push back and really use your brain because chatgpt sometimes it just be saying stuff like Lex was trying to fix it before she enlisted my help and she was like taking the front off. And I was like, you have to take the door off then. And she was like, no. And I was like, I'm looking at it. In order to take that front part off that you're saying you would have to take the door off. And like ChatGPT didn't say that because like it doesn't. It just like puts words together. It wants to be right. Like it's not. It wants to help you out, wants you to stay using it, right? It's just, it doesn't know anything, right? It's a math. [00:24:07] So you also have to push back and be like, wait a minute and use your brain, right? So don't just take the advice at face value. But uh, it can be really helpful. Start and then you can enlist the help of actual people via things like Reddit and especially Reddit's amazing and YouTube. And suddenly your wine fridge is fixed for, you know, 50 bucks plus the cost of a new multimeter instead of paying $500. So that, that is how I use ChatGPT this week. And that, my friends, is all for today. Hopefully you found this episode helpful if you did consider leaving a reading or review. I love them. It's just is they're so fun to read. And yeah, I'm just super grateful. So do that if you want. Don't forget I also have a companion newsletter that drops every Thursday that is basically the podcast in text format. So if you prefer to read or you just want to wreck record, join the newsletter family, you can head to chatgptcurious.com newsletter or just check out the link in the show notes. As always my friends, endlessly endless, um, endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.