Transcription
Well, it's been one of those weeks where it's hard to know what to believe, because it was the week of April Fools. And when you're in a world where AI is sort of like unbelievable enough, now we're getting fake unbelievable stuff.
For example, on April 1st, 11 Labs launched text-to-bark. LTX Studio announced that they acquired OpenAI's Sora and they were planning to open-source it. And OpenAI actually released a new voice inside of ChatGPT that really kind of doesn't want to hear from you. "Oh good, you found me. Uh, yay." And that's just a handful of the April Fool's jokes that came out. So, for the most part, I kind of tuned out everything that came out on April Fools and then the next day sort of double-checked to see what was real or not.
So, as I release this video, I'm actually out at Microsoft's 50th-anniversary event in Seattle. I'm recording most of the news on Wednesday, but you might actually see me jump to my hotel room throughout the video so I can add some updates at the end of the week of any additional news that came out. But I'm not going to waste any more of your time. Let's get into the actual real news that happened this week.
Starting with news that probably has a lot of people excited after seeing the flood of Ghibli-style images. And ideally, after watching my recent video where I showed over 50 ways to use this new ChatGPT model, but ChatGPT actually rolled out the model for free users. Now, now there are some limits. It doesn't actually say in any of the OpenAI news or documents, but Sam Altman did previously say that free-tier users would get three images per day. And when The Verge tested it, that's what they got: three images per day.
And with all of this craziness around the new ChatGPT image generation model and it sort of flooding social media with these AI-generated images, well, ChatGPT had its biggest day ever. Sam Altman took to X to say that ChatGPT, launched 26 months ago, was one of the craziest viral moments I'd ever seen. They added 1 million users in 5 days. Well, after this latest launch, they added 1 million users in 1 hour.
And despite it being available for both paid and free tiers now, Sam Altman is warning that you're probably going to see announcements from OpenAI actually slow down as a result of everybody overloading their servers. Once again, Sam took to X to give the update. He said, "We're getting things under control, but you should expect new releases from OpenAI to be delayed, stuff to break, and for service to sometimes be slow as we deal with capacity challenges." But hopefully, this will help them get that under control because this week OpenAI raised another $40 billion, led by SoftBank. They raised a $300 billion valuation. Hopefully, some of this money is going to the GPUs and data centers to get things under control because of all of the demand that they're dealing with right now.
OpenAI also announced this week that ChatGPT Plus, the $20-a-month plan, is now free for college students in the US and Canada through May. And since we're talking about OpenAI, we might as well mention this as well. They're actually planning to release another open language model. They put out this page on the OpenAI website saying, "We're planning to release our first open language model since GPT-2 in the coming months. We're excited to collaborate with developers, researchers, and the broader community to gather inputs and make this model as useful as possible. If you're interested in joining a feedback session with the OpenAI team, please let us know." And then they posted this form. Sam also shared the post on X, saying, "TLDR; we are excited to release a powerful new open-weight language model with reasoning in the coming months and want to talk to devs about how to make it maximally useful. We're excited to make this a very, very good model." I don't know why this needed a TLDR, because this tweet is about as long as the post itself, but I feel like with all of the open-source models doing better and better and better and Sam's comments a couple of months ago about how he feels they may have taken the wrong approach by not releasing open source, well, they're sort of writing that ship a little bit and planning to put some more stuff out as open models.
And one last thing that OpenAI did this week was they sort of stealthily launched the OpenAI Academy. There was no big fanfare or announcements around this, but if you go to academy.openai.com, you can actually find educational resources on how to use AI, like AI for older adults, automate knowledge graphs with RAG, AI for nonprofits, and much, much more. These all appear to be online training sessions, but they also have this content section here, which is pre-recorded educational stuff.
Quickly, before I transition away from talking about OpenAI, apparently the Chinese model Ernie 4.5 actually beat GPT-4.5 at chess. Ernie 4.5 played ChatGPT in a three-game match here. And apparently, this Ernie model beat GPT-4.5 three out of three times.
Google also made their best model free this week by releasing Gemini 2.5 for all free users. So if you go over to gemini.google.com, google.com, and you click on the dropdown, you should be able to see 2.5 Pro Experimental. And this is a really good model. I've talked about it in previous videos. It's amazing at code. And it's got a massive million-token context window, meaning that you can input and output about 750,000 words, which makes it so great for coding because you could put like massive code bases in here and have it understand the entire codebase.
Google also rolled out some updates to NotebookLM. You no longer have to actually give it a whole bunch of sources that you want it to use to actually create the content around that you're going to chat with and to create your podcasts from. It now has a new discover sources feature. Now, when you tap the discover button in NotebookLM, you can describe the topic you're interested in, and NotebookLM will bring back a curated collection of relevant sources from the web. You can add those sources to your notebook in one click. Now, supposedly, when you log into NotebookLM under sources, there should be a button that says discover sources, according to their instructions here, but I'm not seeing it yet. So it must be sort of a rollout. And well, it hasn't rolled out to me yet. But according to Josh Woodward here, who's the VP of Google Labs, in the last 10 days NotebookLM has rolled out mind maps, discover sources, better PDF understanding, enterprise-grade data protection, and linking back to original sources. So NotebookLM, if you haven't been paying attention to it, they're rolling out features constantly in it lately.
And since we're talking about Google anyway, I want to mention that Google Slides also added Imagine 3 as an option to add images into your slides.
Amazon's getting into the AI agent game as well with their Nova Act. This is Amazon's attempt to take on OpenAI's Operator and Anthropic's Claude, or something like Manis that we've seen. And at the moment, it's available for developers to use over at nova.amazon.com. You can see inside of Nova you've got sort of a normal chat option, uh, generate image area, but down under labs you've got Act. And we can see in their little demo video here that it's thinking, "I should click on the tab." And this one is trying to find the dream apartment. You can see it actually typing in. You can see the thinking on the video here. And then the mouse moves and it clicks on different areas of the browser, selects two bedrooms, one bathroom, and the browser is autonomously doing all of this clicking. And you can actually see the thinking process as it's going around doing all of the thinking.
And while we're on the topic of AI agents, the company Rabbit, who makes the little Rabbit R1 device, is making AI agents that can use your computer as well, but you don't need the little Rabbit device anymore. They're calling this new agent Rabbit OS Intern because they claim it's sort of at the level of what an intern can do right now. Taking a look at their demo video here, we can see that it gives it a task: "Make me a tool that allows me to make 16-bit music loops. I want there to be eight rows of bits where I can place the sounds, etc., etc." It then says it's working on it. And then it sets up step-by-step tasks, very similar to what we saw with Manis. And then it goes through and it's creating all the code files, gives them an index.html file, and there they have the app that this tool went and developed for them.
As of right now, it says they've opened a free trial of the upgraded AI-native operating system. It's available for anyone to try for free on the Rabbit Hole website, hole.rabbit.ext, for a limited time. If you're an R1 owner, you get nine tasks per day, while everyone else gets three. So I'm logged in here to give it a test, and I'm going to go ahead and use one of the example prompts here about analyzing a company's financials. And let's have it analyze Tesla and see if it has the ability to achieve a $10 trillion market cap. And let's go ahead and run it and see what it does.
All right, so it's asking me some clarifying questions. It looks like to analyze Tesla's fundamental performance metrics and evaluate the potential for a $10 trillion market valuation, I need to cover the following areas. If you have any specific data points or sources you want me to include, please let me know. I'm just going to go ahead and tell it to go ahead and start. So now it is starting the plan generation, creating a task: "Gather and analyze Tesla's historical financial data." Task initiated. And if we look over on the left side here, we can see the list of tasks. So it's going to gather and analyze Tesla's historical financial data, analyze Tesla's market share evolution and product portfolio structure, assess Tesla's position within global IT spending and technology sector, evaluate Tesla's potential for reaching a 10 trillion market valuation, compile all analysis into a final comprehensive report. And as we can see, it's going and doing all that right now. Very similar workflow to what we've seen with Manis. It's still working on this first task here. So I might sort of fast forward and let you know the outcome here.
And now it's finally done running. It did take a good 15 minutes. I actually stepped away, did some other things, and then came back and it was still actually completing the task. But you can see it did all the tasks over here, and then it generated all of these various like markdown files here. We can actually go back and look through the whole process of everything it did here. At the end, we finally got this final presentation report. We've got our overview, the final output. It's explaining what the files are here, and then it gives us our key findings. And it looks like it only analyzed data through 2023. We're in 2025 right now. So don't know why it didn't actually get data at least through last year. And then it gives me some instructions on how to use a report along with the report structure. So something fun to play around with a little bit more in the future.
The Rabbit guys did message me and let me know about this update, and they said their goal is to try to bring technology as good as Manis to the United States.
There was an interesting update out of X this week. Apparently, xAI has acquired X, the social media platform. It valued xAI at $80 billion and X at $33 billion. Basically, Elon Musk sold an Elon Musk company to Elon Musk. The idea being that if xAI owns X, the social media platform, then xAI doesn't have to jump through any sort of loopholes or red tape to be able to use the data that is owned by X.
A little bit of news out of Apple this week as well. Some of the Apple Intelligence features expanded to new languages and regions. Their AI features are now available in French, German, Italian, Portuguese, Spanish, Japanese, Korean, Chinese, as well as localized English for Singapore and India. Also, the Apple Intelligence features are now available in the EU. They also brought Apple Intelligence to the Apple Vision Pro this week with their VisionOS 2.4 update. So now if you're, you know, doing things like writing emails inside of your Apple Vision Pro, you have AI features within your email. You've got access to the image playground where you can generate images inside of Apple Vision Pro. You can generate Memoji inside of Apple Vision Pro. There's natural language search within your photos. And a lot of the Apple Intelligence features that they've been pushing out to iPads and iPhones are now available in the Apple Vision Pro.
This week for this video, I partnered with Adobe Firefly to introduce you to something amazing that they've just launched: the new Generate Video, powered by the new Firefly video model. Adobe Firefly offers powerful AI tools specifically for creators who need commercially safe, production-ready content. So what exactly does commercially safe mean? Adobe Firefly is trained exclusively on licensed content like Adobe Stock images and public domain resources. This ensures everything created using Firefly is safe for commercial, professional, or educational use. For creators, knowing your work is responsibly sourced can be a big advantage. So let's dive deeper into the Generate Video module and explore how it can help creators. You can easily produce dynamic 1080p videos simply by using text prompts or reference images. Imagine you're working on a project and need a specific custom B-roll clip, a dramatic drone shot sweeping over a bustling city skyline at sunset, for example. Instead of spending hours searching through stock footage, you can use Firefly to quickly generate the exact clip you need by typing your description and setting parameters like camera angle, motion speed, and lighting. What's especially useful is Firefly's integration with Adobe Creative Cloud apps. Once you generate your custom footage, it can be imported directly into Premiere Pro and placed seamlessly into your editing timeline. Need atmospheric elements like lens flares or smoke? Firefly can generate those too, easily blending them with your original footage when imported into Premiere Pro or After Effects. Another impressive capability is animation creation. Suppose you have sketches or storyboards you'd like to animate. With Firefly, you can upload these reference images, set start and end frames, and generate animated sequences that help visualize your creative concepts. You can further refine these animations using detailed editing tools in After Effects. Adobe's approach to responsible AI innovation emphasizes transparency, accountability, and respect for creators' rights. This means you can confidently integrate AI into your workflow, knowing Adobe actively supports creators and respects copyrights. If you're interested in exploring what Adobe Firefly can do, check out Generate Video in the new Adobe Firefly web app. It's user-friendly, intuitive, and can help enhance your creative projects. Visit firefly.adobe.com and try it for yourself.
Speaking of Adobe, they just rolled out the feature in Premiere Pro where you can use AI to extend videos. This is a really cool feature. It's a feature where, let's say you have some B-roll in your video, but the B-roll's too short by like 1 second. You can actually extend that video on the timeline, and it will actually generate a few extra frames on that video based on what it saw before. If you have sound effects in your video, but the sound effects don't span long enough in your video, you can actually now use AI to generate more sound effects or a longer sound effects track on your video. Some really helpful editing features for people that edit in Adobe Premiere.
There was a ton of AI video updates this week, starting with Runway's new Gen 4 model, which is a really impressive model. I mean, this is like V2 quality video out of Runway, possibly even better than V2. I would say they're pretty comparable right now. In fact, I have a video in the works that I'm going to be releasing soon where I try to compare all of the available video models, and we try to look at which is best at which types of videos. And now that Runway Gen 4 is here, we have a new one to compare. But all of the examples I've seen so far have looked pretty dang impressive. If you have a Runway account, you can log into Runway, click on Generate Video down under the model here, and you should be able to select a Gen 4. Now it looks like Gen 4 is only image-to-video. I can't enter just a straight-up text prompt; it doesn't seem, but I can click, "Create an image," generate an image of, let's say, a wolf howling at the moon. Click generate, and we've got some decent starting images. So let's go ahead and use this top left one here. We've got Gen 4 already set. And if we want, we can describe our shot. Let's just go ahead and put the same prompt: "A wolf howling at the moon." Click generate. And after roughly a minute and a half to 2 minutes-ish, I've got our 5-second video of a wolf howling at the moon. Looks pretty good. I like the extra steam and then the wolf running away. That's some new elements I haven't seen when doing this generation. Let's try a monkey on roller skates and generate that as our starting image here. I think I like this one where the monkey's at the beach. So let's go ahead and use this one for a video. And here's what we got out of that one. The monkey's actually doing some dancing on the roller skates. And it actually got the waves crashing in the background. I was wondering if those would be frozen or actually move. I would say the roller skating is a little bit funky, but I mean, compared to what we were getting a year ago, it is still really dang good. So again, that's Gen 4 available in Runway now. And we will be doing some deeper dive videos on the various AI video models in some future videos on this channel. So make sure you're subscribed for that.
There's a new AI video model in town as well, called Higsfield AI, that came out this week that claims it can do cinematic shots with bullet time and super dollies and robo arms all from a single image. So looking at some examples, we got that like dolly zoom effect. We've got that robo camera effect, uh, some cool panning, some almost like drone-looking shots. I mean, it looks like it can do some pretty cool stuff. Here's an example of a snorry cam. I'm actually not even familiar with that term. A crane over the head shot, some fisheye video, head tracking. That's pretty cool. A whip pan shot, a through-object shot, and it looks like you can check it out over at higsfield.ai. Tons and tons of examples on their homepage. Let's take a peek at the pricing tab here. So they do have a free trial with free generations, limited access, and a watermark. It looks like you get 25 credits, which is, I guess, roughly two video generations. And then for six bucks a month, billed annually, or nine bucks a month, billed monthly, you get about 15 video generations. I just logged in. Let's click on the create tab here. And once again, it looks like you need to start with an image, but it does have a text-to-image generator here. Let's go ahead and prompt a wolf howling at the moon here. I'm not trying anything too complex in this video; I'll save that for a future video. I mean, the images kind of lack some detail, honestly. It's more of like a silhouette, but it looks like it enhanced my prompt for me and actually prompted a silhouette. But let's go ahead and click on video here. And we'll use this image as our start for the video. And let's do like a robotic arm camera that pivots around the wolf. Let's see what happens. What do we have under advanced? All right, we got duration, seed, CFG, and steps. So I'll just leave that all by default. And let's generate and see what this does. Now, as I'm waiting for this to generate, I just realized that if I want to get these various effects that they promised, they actually have a section up here where it says General. And if I click change, you can actually see this is where you would actually set some of the effects. So you've got things like 360 orbit, action run, arc, basketball dunks, buckle up, bullet time, car chasing, car grip, crane down, crane over the head, etc. And if I click show all, you can see all of the options. So there's things like super dolly in, super dolly out, robo arm. I probably should have selected one of those. And that's how you would get one of those effects. So I don't have very high hopes for this generation that's going right now. All right. It took several minutes to generate, but here's what it came back with. And honestly, it's better than I expected.
Luma AI rolled out some new features for their AI video generation. Very similar to what we just saw with things like crane down, crane up, static orbit, left pan, left pan right, pedestal down, pedestal up, pull out, push in, etc. So almost the exact same feature set that was promoted by Higsfield AI is the new features that Luma AI just rolled out. Again, with all these new video features coming out recently, we're going to have to do another video to dive into this. Otherwise, I will end up making 30 minutes of this news video be all about AI video generators. So let's go ahead and move on, and we'll test this in a future video.
Korea AI rolled out some new tools this week, including this 3D tool and a complete overhaul of their website. They also rolled out the ability to edit images with natural language. Though very similar to what we got from ChatGPT this week, you can now chat and get it to edit the images by saying things like, "Put the car on top of a rock." It looks like they saw the success of what ChatGPT did when they rolled out their new O4 features and rolled those into Craiyon using the Gemini language model. Korea also rolled out a new video restyle feature where you can upload videos and restyle it based on any sort of design style that you upload. Very similar to like the Ghibli-fication we've been seeing coming out of ChatGPT. You can do that with videos. If you remember Runway's Gen 1, this seems like a very similar concept to that, just a lot newer, so likely a lot better.
Meta showed off some new research this week with their Mocha towards movie-grade talking character synthesis. And we can see some examples here of what this is capable of. "What are you going to do? I got to follow my own plan. Stay here and get my own thing going." Now the voices to me sound really, really good. I don't really feel like anybody's nailed lip-syncing yet. It still feels a little bit off to me. Let's take a peek at another one here. "I said it before, and I'll say it again. Life moves pretty fast. You don't stop." Visually, it looks cool. Audibly, it sounds cool. But there's just still something off with the lip-syncing. "Sometimes it is the people who no one imagines anything of who do the things." We can see here all talking characters are generated solely from speech and text. So there was no initial video or image to get these kinds of things. "This is Harvey Dent, the famous Bruce Wayne. Rachel's told me everything about you." Now this is something we don't have access to yet. This is just sort of research that Meta showed off. I imagine they'll be rolling it out somewhere at some point, possibly even making it open source.
There's a good chance we're getting Midjourney version 7 very, very soon. According to Midjourney in this post on April 2nd, "We're ending the Relaxathon as we prepare our server clusters for the V7 model launch." In fact, because I'm recording this portion on Wednesday and some of this on Thursday, if it came out on Friday, it may have already rolled out by the time this video has been released. I don't know exactly when they're going to roll it out. There's a very good chance that V7 might be out by the time you're actually watching this video.
11 Labs just rolled out a new feature called Actor Mode where you can now use your own voice to guide the delivery of scripts that are spoken. So let me show you what this sounds like just by default. "To be or not to be, that is the question. Whether 'tis nobler in the mind to suffer the slings and arrows of outrageous fortune, or to take arms against a sea of troubles, and by opposing end them." To use Actor Mode, I'm going to select this text, I'm going to click, "Direct speech with your voice," and then I can upload a previous recording of myself reading this out, or I can record it right now in ElevenLabs Studio. "To be or not to be, that is the question. Whether 'tis nobler in the mind to suffer the slings and
To be, or not to be, that is the question. Whether 'tis nobler in the mind to suffer the slings and arrows of outrageous fortune, or to take arms against a sea of troubles, and by opposing end them. To be, or not to be, that is the question. Whether it is nobler in the mind to suffer the slings and arrows of outrageous fortune, or to take arms against a sea of troubles, and by opposing end them.
Miniaax from Halo AI just rolled out a new speech model as well, where you can turn any file or URL into lifelike audio. So you can create audiobooks, podcasts, things like that, with up to 200,000 characters in a single input. That's pretty crazy. Like full-on audiobooks. Here's the example they shared: "Come, come, come. I see you're using AI tools already. So smart, but h cannot just rely on tools only. Law, the future belongs to those who can work alongside AI, not those scared of it." There's some more examples here in this X thread, which I will link up as I always do in the description below.
And this company, MA, just released a new music generation tool similar to what we've gotten from Suno and Udo. And here's a quick example of a country song. [Music] I said, "I'm getting used to feeling…" I don't know if I'd say it's on the same level as like Suno or Yio yet, but it's really dang close. And here's a jazz song. "I used to shut my door in the…" I mean, again, pretty good, but I do feel like I'm still getting a little bit better generations out of Suno personally.
If you're into AI coding, we got some cool updates this week. Windsurf, the tool that I've tended to use the most recently for coding, dropped some new features. You can deploy your apps directly from Windsurf, similar to how you might use something like Bolt or Lovable. It generates commit messages for you when you commit to like GitHub, and it's got improved MCP support. I'm super excited about these features. I think the biggest feature among this is the ability to deploy an app, so you can generate directly inside of Windsurf and then deploy it directly to a place to manage your front end, like Netlify. So pretty big update there to cater to people who know even less about coding, that have no clue how to deploy an app once they've developed it inside of Windsor.
And Cognition Labs, the company behind Devon, the $200 a month AI coding assistant, they just rolled out a new Devon 2.0 with a $20 option. Now I don't believe it's $20 exactly. It's $20 to get in and then more of like a pay-as-you-go model. I haven't tested this one myself yet, but another coding tool in our toolbox to play around with if you're into the whole vibe coding or using AI to help you with your existing coding.
My buddy Riley Brown is rolling out a Vibe Code app where you can build your own apps directly inside of a mobile app and then deploy and use that app. Something pretty cool to check out. And Riley's a friend of mine, so I want to give him a shout out.
This week, Claude also rolled out Claude for Education, which is taking a different approach to AI where it's not just giving people the answers, it's actually helping them work through solving problems. We can see here there's a new learning mode where they describe it as a new Claude experience that guides students' reasoning process rather than providing answers, helping develop critical thinking skills. This looks like it's going to be made available to various university campuses and should be a pretty decent approach to get more educators on board with using AI and using AI as a sort of assistive learning tool instead of just an answer-giving tool.
Now this one's interesting: Tinder has a new AI-powered game where you try to flirt with this app and then tells you whether or not you're ready to go and flirt with real people. We could see in this screenshot here that they're giving people scores like minus 10 for persistence and plus three for charming, and giving tips as they try to flirt with this AI game version of Tinder.
And a few last little things here. There's a new technology that transforms brain waves to voice, giving speech to the speechless. The brain-to-voice neuroprosthesis works almost simultaneously with the user's intent to speak. It processes brain signals at 80 millisecond chunks, producing speech that flows naturally as the person thinks about forming words. So people who have paralysis that affects their speech, they can now hook up these little brain attachments and be given the ability to speak again.
Apparently, Meta's about to roll out some new smart glasses, like the Ray-Ban Metas, but these ones are going to have like a little screen in one of the eyes with a little heads-up display, and they're planning on selling them for over $1,000. So they're going for a little bit higher-end market than the Ray-Ban Metas, which cost about 300 bucks. These are not the Orion glasses which we saw at Meta Connect last year. Those are still a ways off and still would be insanely expensive if they were released today.
And that's what I got for you today. Some really, really cool stuff happened this week. Again, some of the clips may have been shot from my hotel room in Seattle. I recorded most of this on Wednesday, some of it on Thursday, and if any new news came out on Friday, that'll be included in the next news video that I make.
I have one little request that maybe you can help me with. In this year's 29th annual Webby Awards, the podcast that I co-host with Nathan Land, The Next Wave, is up for a Webby award. We're up against some really, really mega podcasts. So it would be awesome to try to win this thing. So if you listen to The Next Wave podcast and you enjoy it, we could really, really use the vote. This is actually voted on by people who listen. I will link it up in the description. And if you could help us out by voting for The Next Wave, that would be absolutely amazing.
But again, that's all I got for you today. Hopefully you enjoyed this video. If you like videos like this and you want to stay looped in on the latest news and get cool AI tutorials, make sure you like this video and subscribe to this channel, and I will make sure that more stuff shows up like this in your YouTube feed.
And if you haven't already, check out futuretools.io. This is where I curate all of the coolest AI tools that I come across, share the latest news, and I've got a free newsletter that we send out twice a week with just the most important news and coolest tools we come across. Also, if you sign up, you'll get free access to the AI income database, a list of cool ways to make money with AI tools. I think you'll really dig it, and it's all free over at futuretools.io.
Thank you once again for tuning in. I really, really appreciate you, and it's always fun to nerd out. And thank you so much to Adobe for sponsoring this one. I will hopefully see you in the next one. Bye-bye.