AI News: Is OpenAI Speed Running Their Downfall?

There was an absolute ton of news in the world of AI this week, and I want to try to break it all down for you and I do not want to waste your time. So let's go ahead and get right into it. Let's start with some OpenAI news. This is actually news that came out last week, but I didn't get to it in last week's video. And that's that they're rolling out a new plan called ChatGPT GO. Now, ChatGPT GO was already rolled out in India. It was a much lower cost plan. To be able to use chat GPT, you get 10 times more messages, file uploads and image creation than you would if you were using the free tier. And it cost $8 a instead of what you pay for the plus plan at $20 a month. There's also a longer memory and context window, so you can give it a lot more text and it will remember a lot more of the text that you gave it. Now, there's a lot of discourse going around right now about OpenAI and what's happening over there because people think they're struggling. They're rolling out cheaper plans to try to entice more people to, you know, pay them money. And they also announced that they will be doing advertising both in the free tier and in the Go$8 a month tier. In fact, they also put out an entire their approach to advertising. And a lot of people are basically saying OpenAI's kind of speedrunning their downfall. Right now. OpenAI's bleeding at a rate of billions of dollars a year. So they need to figure out ways to generate more revenue. And their best course of action right now is a cheaper plan because not many people are paying for the $20 a month or $200 a month plans. And to, you know, throw in advertising and follow the sort of meta Google model. Now, they do say ads do not influence the answers that chat gives you. So when you give it a prompt, like asking a question, the response isn't going to include advertising. The advertising will be sort of outside of the chat, but it will be contextual to what you're chatting about. None of your conversations go to the advertiser. So advertisers are never going to actually see your chats, but those chats are going to inform OpenAI on what to actually advertise to you. And then we have Google, who famously makes most of their money off of advertising, saying that they really have no plans to put ads inside of their chatbots. I mean, they don't really need to. Their advertising business in search and YouTube and all of the other places is sort of subsidizing the whole Gemini AI thing. Again, this is all really fascinating to me because, well, people are probably going to go and use the chatbot that A gives them the best answers, but B has the best user experience. It almost feels like the large language models underneath are all getting so good that they're kind of like a commodity. And if they're all as good as each other, I feel like most people are gonna go use the option that's the least expensive and also has the best user experience, which is at the moment, you know, seeming like it's less and less going to be chatgpt. And if you want even more interesting news out of OpenAI and the seemingly speed running the downfall, apparently they're planning to take a cut of customers AI aided discoveries. So what does that mean? Well, here's what The CFO of OpenAI said her company could for instance, take license to the drugs that it discovered. So if you're using an AI model and that model helps you discover a brand new drug, well, OpenAI wants a piece of that. In an article the CFO also said licensing, IP based agreements and outcome based pricing will share in the value created. Now, to be fair to OpenAI, they're not the only one who is thinking about doing this. They just kind of have been the most public about it recently. According to this article on the information here, it says it's rivals, anthropic Google, DeepMind and isomorphic labs, which is, you know, from Google, are focusing on using AI for drug discovery and have also held discussions with early stage biotechnology startups about data licensing and partnerships. So they're all kind of looking at this. But right now the sort of narrative is that OpenAI is starting to fall behind in the race. Like their models are sort of all relatively similar now. Anthropics and Google's and OpenAI's models and even GROK are all fairly similar in capabilities now. But OpenAI is the only one rolling out ads and sort of out there publicly talking about how they want to license sort of opportunities that come as a result of their bot. I don't know, it's all just super fascinating to me and I'm just curious to see how it's all going to play out for OpenAI. And since we're talking about OpenAI, might as well throw this in there as well. This week they actually started rolling out the age prediction on ChatGPT to help determine when an account belongs to someone under 18. Now, this is something they've been talking about for a while. They even put out articles weeks and weeks ago about this. But it appears that it's finally rolling out. It's going to be pretty interesting because a lot of us have the mentality of children sometimes, so it's gonna be fun to see if it thinks some of us are actually kids when we're not. We'll see. We've got some interesting advancements in the world of AI video. This week, Runway introduced image to video for Gen4.5. I guess in the past, it was only text to video, so now image to video is an option. But if we jump into our Runway account here, we can see we're gen 4.5, and we now have the option to drop an image in. I'm not going to do a whole bunch of testing in this video, but I will drop one video in. This is just like a screenshot of me holding, like, a toy lightsaber here, and let's just say an epic lightsaber battle in the office and see what it gives us. So that took maybe two minutes or so. Let's see what we got. That's. I mean, I was going to say that's not too bad, but there was some wonkiness going on there. This part through here looks really, really good, actually. But then there's a moment where the lightsaber goes behind my back, but my arm also somehow goes behind my back, and that's where it got a little wonky. And then I change into a different person here, but I'm actually pretty impressed. That's. That's pretty fun. That worked. What am I doing? Oh, that's awesome. All right. I could play with this a lot more, but like I mentioned, I have a lot to cover, so let's move on. That One is gen 4.5 image to video. Since we're talking about AI video, we got some news out of LTX Studio as well. They introduced audio to video in LTX, and they partnered with 11 Labs to make it happen. You essentially upload an audio file. I'm assuming you either give it an image or describe what you want the video to look like, and it uses the audio along with the video. So here's their demo. You probably weren't expecting me to sound like this, but that's the beautiful thing about innovation. You think you've heard it all before, every cliffhanger and every twist, every revolution to how you work. But then, I mean, yeah, it's not bad. And they were right. I was not expecting the man's voice to come out of the woman driving. So it says upload audio, add an image and a prompt or generate straight from a prompt and produce video with native lip sync. To this day, I have yet to find a lip sync tool that I'm really, really enjoy impressed with. Now it works okay when it's like an animation, you know, claymation and weird stuff like that, but for the most part it never feels right. It never feels like it's perfectly synced up and I don't feel like this one's a whole lot better. I am a fan of LTX Studio. I just I'm still waiting for a lip sync tool that I'm really impressed with. While I'm researching for all these AI news videos, I often run into the classic clickbaity AI productivity hack posts, so I wanted to share one that's actually real. What a lot of businesses don't know is that they're usually sitting on a gold mine of data, but it's trapped inside of PDFs and contracts and forms and documents that no one has time to manually dig through. But a new tool from Box called Box Extract uses AI to securely extract the exact information you need and turn it into metadata that's actually useful to your business. It's different from a normal LLM because it's built to actually understand your unique documents like contracts, product specifications, policy documents, and charts. And it uses agentic AI to pull out the most valuable insights, even if it's in a bunch of different places and formats. Box Extract combines the best AI models from Google, Anthropic and OpenAI with advanced agentic capabilities and boxes, security and compliance checks so you don't need to worry about using it for your business. It helps put your data to use so your documents don't have to just be static files anymore, and instead can help you quickly make decisions and automate workflows. If you want to see how Box Extract can make your business more productive, learn more at the link in the description. Thank you so much to Box for sponsoring this portion of today's video. Now let's get back to it. Let's move on to Text two Image models, because there's a little bit of news that came out of that world this week as well. The company Black forest Labs released Flux2Klein this week, and this is a model that you can actually run locally with hardware as little as 13 gigabytes of VRAM. Now we've obviously seen a lot of text to image models. They've been around for what feels like forever at this point. I've been talking about them for almost five years. Yeah, over five years now I've been talking about text to image models. So this is not new technology. So what makes this one new? Well, you can run it locally on your computer, like I said, and when you do run it, it can generate images in less than one second. It can create very photorealistic images and it's got open weights, so you'll be able to customize and fine tune it. Now, they do have a demo that you can use. If I scroll down here under resources, they do have a demo. I'm going to do a superhero monkey kicking a soccer ball at the moon while eating a taco. Yeah. And their demo is definitely not sub 1 second generation, so you need to be running it locally. Since this is going to the cloud, there is some lag, some latency for it to send it to their cloud services and then they do a bunch of safety checks on it and stuff like that. So when you're using their demo, this is not a good example of how fast it can be. But let's see how it looks. Still pretty quick. It didn't get the superhero part, but it's definitely a monkey kicking a soccer ball at the moon with a taco. But now that it's already sort of booted up and loaded one image, every time I do it from this point on, it's a little bit faster. So if I hit this regenerate button down here, you can see that it's starting to generate some new images. And that was all in real time. You can see it just generated six images and I did not speed that up at all. I really hope that's a tale. The company Korea AI this week put out a demo video for their Real Time edit. You can get on the wait list and join the beta. I haven't gotten access to this one yet. I'll definitely play around with it once I do. But just looking at their demo, you can kind of see what it does here. So they have an image of a car here and down in the bottom you can see the prompt turn into the car buried with coins. And then it adjusts it and then it's underwater. And you can see as they move this around, it updates in real time on the image on the right. And as they type the prompt, it actually updates the image on the right too in real time while they're typing. So looks pretty cool. They can even put it into like A camera mode and have it generating images in real time directly from the camera. So it looks like a fun little toy. I'm not quite sure how useful it's going to be, but I guess you could sort of dial in the image exactly the way you want it and really, really fine tune it by seeing it as it's generating. So that's kind of cool. All right, so from image to video, then audio to video, then text to image, then time image generation. Now let's talk about text to speech. Everybody knows about 11 labs. It's a great text to speech model. Well, Alibaba this week released the Quinn 3 TTS model, or text to speech model. This is actually an open source model that you can run locally on your computer and it's pretty fast. And I have tested this a little bit. You can test it over on hugging face. I'll link up this tweet below. But if you go to this hugging face link here in the tweet and then down here you can see there's a Quinn 3 TTS demo. So coming in here, we can see that we've got voice design, voice clone and text to speech, custom voice. So here's the example that they have here where you can see this is the text they inputted. But then you actually describe what you want the voice to sound like. Speak in an incredulous tone but with a hint of panic beginning to creep into your voice. Let's generate with custom voice. Going to take about 20 seconds. And here's what that sounds like. It's in the top drawer.

Wait, it's empty.

No way. That's impossible. I'm sure I put it there. That was a lot of panting in the middle of there. I don't know what that was about. Let's try a variation. I'll do an angry male screams in anger. I know that's redundant, but I just want to see what it does. And here's what we get with the new voice description. It's in the top drawer. Wait, it's empty. No way. That's impossible. I'm sure I put it there. So pretty cool that you can actually design sort of the emotion behind the voice. They also have this voice clone feature here, which I actually thought was pretty impressive the first time I tried it. But basically you give it some audio of the voice you want to clone. You give it the text of exactly what you just said and then you give it some new text and it will generate new audio, cloning that voice. So let me switch it to the Microphone here. And then I will record just a real quick snippet of me saying something. My name is Matt Wolf and you are watching my YouTube channel. Don't forget to like and subscribe. I then type in the exact text that I just said in this audio sample and then I give it some new text. I'll click clone and generate once again. Takes about 18 seconds. And here's what it sounds like. Let me know if you think it sounds like me. This is the best YouTube channel for people who love AI. It's also the best YouTube channel for people who hate AI. I think it's pretty good for a free open source model. I mean, it's sort of more of a monotone version of me. There's not a lot of inflection in there, but I think it actually sounds like me. And then there's one last thing I want to show you here that I thought was pretty cool. You can actually use Claude code and the various agent tools that are out there to create videos Now. There's a tool called Remotion, which is a Claude skill. You literally open up Claude code, give it this little snippet of code here and it will install the skill and then you can just prompt a video idea. So here's an example of something I actually made with this Remotion inside of Claude code here. I give it the prompt create a 60 frame per second video with a chart that's going up and to the right. The chart should animate to the right, but not be a completely straight line. Think of a stock chart for a stock that's increasing. It goes up and to the right, but not in a perfect straight line. Make it look beautiful and have dollar signs and money animating out of the line chart as it animates upwards. And here's look at the animation that it created for me. I'll press play here and you can see it created this stock chart that's moving up and to the right with little like dollar signs and coins coming out of it. I gave it some additional prompts to add my name here, but this part over here where it's showing the money and the to the moon. AI decided to do that itself. I didn't tell it what to put there. I did some other tests. I literally gave it the prompt make something impressive. I just wanted to see what it would make. This is what was impressive that it made in Remotion here. Beyond the stars lies something infinite. So that was, that was the impressive thing that it made for me. I gave it a prompt to make a kinetic typography that says follow Mr. E Flo and I told it to make it like blue and purple cyberpunk and this was the animation that made not bad. But then I tried to make it do a waving person and well, it got a little wonky. It's not horrible, but somehow his hair goes beyond they hind his eyes and I think his hand's broken in the wrong direction. Anyway, it's pretty impressive for just typing a prompt and getting like a motion graphic style video, even though for objects like this it's still not very good. All right, those were the main things I wanted to talk to you about, but I wasn't lying. There's a lot more to talk about and well, let's just go into a rapid fire and rattle off all the rest. If you're a Spotify customer, you got some new AI features this week. Specifically, you got the prompted Playlist feature. You could give it prompts like mood boosting, indie pop for getting ready to go out with friends, including songs recently featured on hit competition and reality show that fit my taste, along with songs I love but haven't played this year. Or you can begin with a scenario like warm acoustic songs for a slow Sunday morning, plus hidden gems and covers from singers songwriters trending globally right now. So you've been able to prompt playlists in the past, but now it's much more granular, as you can see. You can pick things from like popular TV shows and stuff like that, which just gives it a whole lot more power and helps you really, really dial in that exact playlist that you want. Pretty cool. Google's Gemini rolled out a feature that helps students prep for the SAT with actual practice tests inside of Gemini. So if you're in that demographic where you're getting close to taking the SATs, well, you now have a new resource to really, really help you prep and dial in for the test. And since we're talking about Google, well, they're rolling out personal intelligence in AI mode. We talked about personal intelligence in one of the news videos a couple weeks ago, where the AI has access to your Gmail, your Google Calendar, your Google Photos, your Google Drive. All of your various Google ecosystem products are available, and Google has access to that information when you're searching Gemini before you had to use it inside of the Gemini app specifically. But if you're a pro or ultra user for Google, it is now rolling out inside of AI mode for Google Search. So just another place where you can go and use that Google Personal Intelligence feature Moving on to YouTube news, here this week, the CEO Neal Mohan put out a article about what's coming in 2026 for YouTube. But I'm going to go ahead and scroll down here, all the way down here to point number four. 4. Supercharging and safeguarding creativity. For years, AI has been the quiet engine behind our most important innovations, like recommending the next video or helping to keep the violative content off the platform. They announced that this year you'll be able to create a short using your own likeness. You'll also be able to produce games from simple text prompts, which I've talked about and showed off in past videos, and experiment with AI music. So, yeah, it sounds like they're going to pool a Sora. We all saw that Sora app where you can train it on your likeness and then make videos of yourself doing wacky, outrageous things. Well, that apparently is coming to YouTube shorts, and you're gonna be able to do silly memes like that. Personally, not a feature I'm super excited about, A, because I don't personally watch a ton of shorts, but also B, like, I don't really want to see more random AI avatar videos. But that's just me. I could be proven wrong. There could be some content that people make that I really, really enjoy from this. But just, you know, my first gut instinct is, is I don't know if I'm a fan of that. They also go on to say it's becoming harder and harder to detect what's real and what's AI generated. So they're gonna make sure they clearly label content created by YouTube's AI products. And creators must disclose when they've created realistic, altered or synthetic content. I mean, right now there's already a checkbox in the back end of YouTube to check to say that your content was AI generated. But like, most of the people that are making AI generated content aren't checking that box. So I guess maybe they're gonna get stricter about that. When it comes to managing AI slop, we allow for a broad range of free expression while ensuring YouTube remains a place where people feel good spending their time. Over the past 20 years, we've learned not to impose any preconceived notions on the creator ecosystem. To reduce the spread of low quality AI content, we're actively building on our established systems that have been very successful in combating spam and clickbait and reducing the spread of low quality repetitive content. In my mind, this isn't really saying anything. It's saying we know, there's AI slot, but we're gonna keep doing what we're doing. I don't know if that's actually what they're saying, but that's, that's, that's kind of what that messaging feels like. But I do like their stand. I don't think they should just blanket, say no AI content. That's stupid. Like AI is going to be in everything. Those lines are going to get more and more blurred. So like trying to ban AI is just going to be a cat and mouse game that YouTube's never going to be able to keep up with. Right. People are just going to figure out ways around and better ways to fake it. So it's just too much of a battle. So I don't think YouTube wants to fight it. Moving along to Adobe News, there's some new AI features in various Adobe products. There's some new video editing tools in Premiere, as well as motion design upgrades in After Effects. So it looks like in Premiere here you're going to be able to create masks much better with a lot of motion going on. So their AI is getting better at masking and doing this sort of rotoscoping kind of thing. They're making it easier to bring in media that you've made with Firefly and some other features. But those are the main AI features. So if you're here to learn about where the AI is going and what products it's landing in, those are the AI features that are landing in Premiere and After Effects. But I'm not done with Adobe yet because they're rolling out a new Adobe Acrobat feature that will turn your PDFs into podcasts. So this sounds a lot like Notebook LM to me, where you feed it a whole bunch of information and then it turns that information into a podcast. Except now you're going to be able to give Adobe Acrobat a PDF file and it will turn it into like a quick highlights podcast. In self driving news, the first ever Tesla Robo Taxi is now driving in Austin autonomously. So like a Waymo, you can actually get into the backseat and this Tesla will now drive you around. Now some of these full self driving features have been in Teslas for a while, but this seems to be the first time that one of them will come and just sort of pick you up and drive you around. And speaking of Waymos, if you're in the Miami area, well now you're getting Waymo as well. This week AMD announced a new mini supercomputer called the Amd Ryzen AI Halo it looks like a DJX Spark from Nvidia, and I think it's pretty much designed to be the same thing. It's like a little box that you can put on your desk and it will do the AI inference, like right at home without needing to connect to the Internet. If you're familiar with Nvidia's DGX Spark, this seems to be like the direct competitor to that. Speaking of hardware, it looks like Apple is getting into the hardware game themselves. I think this is kind of more rumors and speculation, but apparently Apple's developing an AI wearable pin that is likely not an actual image of it. They say it's going to be the size of an airtag. It will be equipped with multiple cameras, a speaker, a microphone, and wireless charging. I don't think a ton more is known about this, but it looks like this year could be a year where we get quite a bit more wearables. And on that note, rumor has it OpenAI is on track to launch a physical device this year. They're still being pretty hush hush about what this device is. There was rumors floating around that was going to be some sort of AI Pennsylvania, like not a pin pin that you pin to your shirt, but like a pen, like this kind of pen. But I don't know if that was true. That was just some like random rumor that I don't know where it got started. According to this article, the device would sit on your desk or travel in your pocket, quietly absorbing context and answering questions like a ChatGPT powered companion. It looks like we can expect it in the second half of 2026, but nobody has given any details about what the device is, what it will look like or what it will do. So yeah, we don't really have much to work off other than like it's coming prom Microsoft this week released a new model that's going to go into robots. This week they announced Rho Alpha, which is their first robotic model derived from the Microsoft Phi series. And it's capable of translating natural language commands into control signals for robotic systems performing bimanual manipulation tasks. So essentially, like, you know how we have prompt to image, prompt to video, prompt to answer. Well, this is like prompt to robotic motion, right? So like they're giving it prompts here. Pull out the red wire and then the robot knows to look for the red wire and then eventually it will. Eventually, eventually it will. Yeah, eventually. Yeah, there it goes. There it goes. It will pull out the red wire. All right, a couple more quick stories. Here, Google did another one of these weird aqua Hire sort of things where they managed to gobble up all the talent from a company without actually having to acquire the company to skirt regulations. And yeah, a lot of companies are doing this lately and I still think it kind of tends to screw over a lot of the employees that don't end up moving over to the new company. But this time, Google nabbed up the AI voice startup Hume AI. Hume AI CEO Alan Cohen is joining Google DeepMind along with several top engineers as part of a major licensing deal. If that sounds familiar, it's because we saw Google do that with Windsurf, we saw Nvidia do that with Grof. We've been seeing a lot of companies use this exact play to get the talent and IP from some of these companies without actually having to buy the company. Hume AI is a pretty good voice startup. A lot of people were really impressed with it when I played around with it. It never really lived up to the hype for me. But it's one of the more sort of emotive voices. Like it's got a lot of inflection and tries to have a lot more emotion in the voice. I'm sure the team behind it is a really, really impressive team. So we'll just have to wait and see what Google and DeepMind do. Technology and finally, I want to end on a little bit of an optimistic take from Jonathan Ross. This video here has been going viral recently from this Big Brain AIX account, which is kind of frustrating because, like, they watermarked it as their own video, but it actually comes from this interview with Harry Stebbings here. So let me actually play the clip from the actual original source because the original Source only has 220,000 views and this person here has 1.3 million and gave zero credit to original creator. But here's Jonathan Ross's thoughts on the labor market in the near future.

I believe that AI is going to cause massive labor shortages. Yeah, I don't think we're going to have enough people to fill all the jobs that are going to be created. There's three things that are going to happen because of AI. The first is massive deflationary pressure. This cup of coffee is going to cost less. Your housing is going to cost less. Everything is going to cost less. Which means people are going to need less money because you're going to have robots that are going to be farming the coffee more efficiently. Efficiently. You're going to have better supply chain management. It's just going to be across the entire supply chain. You're going to be able to genetically engineer the coffee so that you get more of it per watt of sunlight just across the entire spectrum. So you're going to have massive deflationary pressure. And what that means is people will need to work less.

So yeah, super interesting take. Basically, he's saying that if everything gets easier to make and cheaper to make, well, it's going to cost a lot less money for people and they're not going to have to work as much, which could lead to labor shortages. I find that to be a really, really fascinating take. I'd be curious what your thoughts are on that in the comments, but that's what I got for you today. Again, a lot of information if you've made it this far, or as Colin and Samir would call it, to the deep end. Well, you've probably noticed that the style of this video has changed up from last several weeks of news videos, and that's because I'm kind of going back to my old ways. I'm trying to get these videos out on Fridays again instead of Saturdays, and I'm trying to break down as much news news and make it as valuable as possible. So you are 100 completely looped in on all the AI news. If that's something you're interested in and you want to stay completely looped in, make sure you like this video. Subscribe to this channel. I'll make sure more videos like this show up in your YouTube feed and you will always be tapped in on the latest AI news. Also, if you haven't already, make sure to check out FutureTools IO. That's where I curate all of the coolest AI tools that I come across, as well as share all of the AI news in real time as I find it. It that's all over at Future Tools IO. Thank you so much for tuning in. I really, really appreciate you hanging out with me today, nerding out with me today, and hopefully I'll see you in the next one. Bye Bye.