AI News: GPT-4o Voice Competitor, Gen-3, ElevenLabs (and More)

Here's all the news from the AI world that I think you'll find interesting from this week. Now, there wasn't a ton of major announcements this week. That's because of the fourth of July here in the United States. I believe a lot of companies sort of slow down on the announcements and big releases during holiday weeks because of travel and things like that. But that doesn't mean there's not a lot to talk about. A lot of really cool things still happen this week, starting with the fact that Gen 3 access was made publicly available. Now, I made a video earlier this week, on Sunday before Gen3 was released, and then on Monday, the day after that video came out, Gen 3 was made publicly available. Unfortunately, Gen 3 is not available. If you're on like a free plan for Runway, you do have to be a pro user. But if you are a pro user of Runway, you can come over to generate videos. You'll see a big introducing Gen 3 Alpha. You can click Get Started and enter your prompt here. I've made a ton of videos with Gen 3 already, and again I. I already made another video about it. So I'm not going to go into too much depth in this video about Gen3 because that whole video will break down what you can expect. However, I did attempt to make a Fourth of July themed video with the prompt a bald eagle flying in front of an American flag with fireworks in the background. And well, this is what it gave me. Yeah, it's not always the best, but it is the best text to video generator we have right now. Compare that to something like Luma AI or where I was able to actually give the prompt to an image generator, upload my image into Luma and I got something that's a little bit more impressive out of it. So Text to video Gen3 is the best we've got today. Image to video, Luma still the best in my opinion. You can't do image to video yet with Gen3.11 Labs got a couple updates this week as well. Last week we talked about their new reader app. This week they've added some famous voices to the reader app, like Judy Garland, James Dean, Burt Reynolds and Sir Laurence Olivier. And before people start freaking out too much, they did get permission from the estates. They made deals. The estates are getting paid for them to be able to do this. So this was all done above board with permission. If you want to check out the app yourself, search out 11 labs on your phone. Make sure that you're downloading the proper version that is created by eleven Labs Inc. A lot of people on the App Store are going to try to fool you into thinking their app is the actual 11 Labs app, but you want the one that actually says 11 Labs. Once you're in the app, you can actually see some of these new iconic voices that are in there. I think it's cool that they have them, but personally, due to the time that a lot of these people were alive and their voices were available, audio quality isn't my favorite. For example, here's what Judy Garland sounds like. Once upon a time in a land far, far away, there was a beautiful young girl who had become lost while traveling. Burt Reynolds is probably the best sounding one just because he's a little bit more recent, a little more current for the good old American life, for the money, for the glory and for the fun. Mostly for the money. I could see listening to articles and PDFs and stuff with Burt Reynolds voice. Eleven Labs also released a new Voice Isolator feature where you can upload any audio with any sort of background noise and it will clean it up and make it sound really good. I particularly like this demo that they put online that shows how good it actually works. Check this out. Need to remove background noise from your video? Use our new voice isolation model for crystal clear audio every time. I mean, it's pretty impressive if you're a fan of Suno and making music. With Suno, they also released an app this week. It pretty much does the exact same thing as the web app. It just makes it easier to do it on your phone. At the moment I believe it's only available on iOS, so if you're on Android, it's probably coming soon. But at the moment you're kind of out of luck when you search it out on the App Store. Search out Suno once again, be careful. A lot of people will try to fool you into thinking their app is the official Suno app. There's one here that says Suno AI by sharp forks. It's not Suno, but there are quite a few apps that claim to be Suno, so make sure you are grabbing the right one. It has this little tie dye, psychedelic looking background and the creator of the app is Suno Inc. So look at the creator, make sure it says Suno Inc. And that's the proper app. Once you're in Suno, it works pretty much like the web version. You've got your library of all the songs that you've created. You've got a create button. Your newest songs pop up towards the top on the web App, your newest songs are at the bottom of the list. Once again, I was trying to generate some stuff for a fourth of July theme. This is what it came up with. You get the idea. Meta released some new research this week around text to 3D images. It's called 3Djin and you can apparently input a text prompt and output a fairly decent quality 3D image. This could speed up things like game development and creating assets for 3D videos and things like that. Here's the demo that they posted on X to kind of show off what it's capable of. We got a little like 3D pug here made out of metal, a dancing robot. It looks like they fed this into Mixamo or something to get that dance and it looks to be pretty good. I mean, I can't really say if this looks a lot better than a lot of the text to 3D we've already seen or if it's about on par and we don't have access to it yet. They just released the research, but once I can play with it, I'm definitely going to play with it and see what we can do with it. This week this company called Kyotai. I'm not sure if I'm pronouncing that right or not, but they're an open source AI research lab. They released a new voice model that looks to be trying to go head to head with GPT4O's advanced voice. And not only is it available right now to try out, they're actually open sourcing it so that other companies could build off of it and use the underlying tech to build new tools from it. If you want to try it out, you can head over to Moshi Chat. It says enter your email to join the queue. I don't know why it does that, because as soon as you click join queue it just spits you right into it where you can use it.

Hello, how can I help you?

What's your name?

You're Moshi.

No, you're Moshi. Who made you?

I was created by Q Tie, a non profit research lab based in Paris.

What things are you best at?

I'm the voice artificial intelligence. So I can understand and respond to human speech.

Are you good at math?

Yes, I can perform basic math calculations.

What's 700,532 times 7?

7532 times 7 is 53. 2303.

That wasn't exactly what I asked, but still pretty impressive.

Thank you.

Now, not quite on par with what we saw out of GPT4. Oh, the voice is not very expressive, yet still sounds robotic. However, the responses you heard were in real time. I didn't edit this to make it seem like it was responding to me quicker than it actually was. It actually instantly responds, sometimes almost talking over you before you finish your sentence. And because this is open source, this is just the very sort of base level, the foundation of what we're going to see from this tech. As soon as other people start using this, integrating it with tools like 11 labs with more realistic voices, integrating it with other large language models that are probably smarter than whatever it's powered by right now, this is going to actually get a lot better and be a decent alternative to GPT4O with the advanced voice assistant. And speaking of open source, there's a new open source large language model that was just made available over on Hugging face called intern LM 2.5. What's interesting about this one is not only that it's open source and but it has a 1 million context window now. Gemini from Google now has a 2 million context window which is probably overkill for most. Let's be honest, 1 million context is probably overkill for most. But now we have an open source 1 million context window that anybody can build with and it's available right now for anybody to play around with over on Hugging Face. The weights are available right now and most likely either now or pretty soon you'll be able to use it in a tool like LM Studio or Jan or ChatRTX. It's kind of like a bring your own model chat system. Speaking of bring your own model chat system, the Brave browser just made an update where you can bring your own model into the Brave browser. So Brave has their own AI similar to what you see from Microsoft Copilot built into the browser. They call it LEO AI and now you can actually use custom models. They've already got Mixtral, Claude and Llama, but now you can actually add a custom model in if that's something you're interested in. Perplexity also got an update this week with their Pro Search. The new Pro Search has multi step reasoning, so it actually understands when a question, requires planning, works through goals step by step, and synthesizes in depth answers with greater efficiency. It can also analyze search results and take intelligent actions based on findings. It's also much better at math and programming because of the addition of Wolfram Alpha. If you're a free user of Perplexity, you can use Pro Search five times every four hours. And if you're a Perplexity Pro member. You get nearly unlimited daily access, so there's some limits, but you're probably not going to find them. I feel like OpenAI and chat GPT really led the way in making chatbots and large language models popular. But if I'm being totally honest, I've become much more of a fan of Claude and Perplexity. I use Claude for most things, except for when I need to do research. Perplexity seems to be the best option right now. It looks like Apple is actually getting a board seat on OpenAI's board. It's going to be an observer role, so they won't have voting rights. But it's super interesting because obviously Microsoft and Apple are like the two largest companies in the world and they're kind of head to head competitors and both companies are actually going to have board roles in OpenAI. So pretty fascinating. But when it comes to OpenAI, another week, another lawsuit. The center for Investigative reporting is suing OpenAI and Microsoft once again claiming copyright infringement. OpenAI and Microsoft started vacuuming up our stories to make their product more powerful, but they never asked for permission or offered compensation, unlike other organizations that license our material. If you remember, recently OpenAI has been signing a ton of licensing deals. Associated Press, Axel Springer, Financial Times, News Corp, Vox Media, the Atlantic, Time makes me wonder if some of these lawsuits are just to try to get on the radar of OpenAI so that it can open discussions to get a licensing deal. At the end of the day, this is all about money for these companies and I'm sure if OpenAI made them a licensing offer that they couldn't refuse, they would get on board with OpenAI pretty quickly. But that's just speculation. Mustafa Solomon, who in my opinion is usually pretty calculated with his words and a very, very smart person, he wrote the book called the Coming Wave, which is a great book on sort of where AI is now and where it's headed. But he made some statements that even I really, really can't get behind with.

Respect to content that is already on the open web. The social contract of that content since the 90s has been that it is fair use. Anyone can copy it, recreate with it, reproduce with it. That has been freeware, if you like. That's been the understanding. There's a separate category where a website or a publisher or news organization had explicitly said do not scrape or crawl me for any other reason than indexing me so that other people can find that content. That's a gray area and I think that's going to work. Its Way through the court.

Yeah, I don't know about that. If I put up content on the Internet, on YouTube, on my personal blog, just because I didn't put like a specific robots txt file telling not to scrape on this, I should just accept that these companies are allowed to use this content however they want. I don't know, I have a hard time getting behind that. I kind of understand what he's saying. If you're putting it on the Internet, everybody can see it anyway. So what's the difference between them seeing it there or seeing it here? I feel like that's sort of the point he's trying to make. That doesn't completely negate copyright. I don't know. What are your thoughts? Do you agree with Mustafa here? Oh, and by the way, if you absolutely don't want to be scraped and you don't want your content taken by AI, cloudflare rolled out a solution this week. If you're on Cloudflare, which is a content delivery network that sort of lives between like your domain and your hosting and it like caches a lot of the stuff on your website to make your website load faster. And I'm not the best person to be explaining this, but it's a service that sort of helps your site load better, helps with security on the site, helps with preventing more downtime on your site, things like that. Now they have a switch that you can flip on that makes it so these scrapers that are scraping for AI can't scrape your site anymore. And it's available to both free and paid users of cloudflare. If you're worried about companies like Microsoft just taking your content and using it however you want, Cloudflare could be a good option for you, especially since they have free options. And speaking of using your content to train AI, last week Figma had their config conference where they showed off all their AI features. This week they put out an official statement to users that says the company needs to train models that better understand design concepts and patterns and figma's internal formats and structure through Figma content. Apparently they are going to offer a way to opt out, but by default, apparently they'll just train on whatever designs and things you make inside of figma. Figma also got into a little bit of hot water this week when it came out that the weather app that it was designing for people looked identical to Apple's weather app. The CEO of Figma did speak up and say that they were using off the shelf language models for their built in AI. So it wasn't specifically trained on Apple's designs, but maybe the off the shelf AI they were using was trained on Apple's designs and they've since paused this feature. So the ability to actually make this kind of stuff in figma is not available right now. While they fix these issues and make sure that it's not going to accidentally create designs that look exactly like other company's designs. YouTube rolled out a new feature where if a YouTuber creates content that simulates what you look like or simulates your voice, you can actually request to have that taken down. Now before, it actually had to be like stolen content. Like if somebody took my videos and put them on their YouTube channel, I can force them to take it down, or if they used my music or my ip, I can claim copyright on it and have it taken down so well. Now if somebody just uses an AI version of your voice or an AI version of your face, you can request it to get taken down as well. Instagram also made a little bit of a tweak. A lot of people were having a bit of a fit about the fact that images that used any bit of AI at all, like if you used Photoshop, it would probably say made with AI on your image. And people were going, hey, this image wasn't made with AI. I just did some color correction that maybe might have used a teeny bit of AI. Or I just erased a little thing in the background and that might have used a little bit of AI. That doesn't mean my image was made with AI. It just means I did a tiny tweak to the image to make it look a little bit better. Well, they changed their wording, so now it no longer says made with AI. Now it says AI info. You can click on it and see what's in the metadata. We also got word this week that a new version of GROK is coming out. Grok 2 is apparently arriving in August. This comes from Elon himself. He was replying to Beth Jesus on X who said models training on each other's data is like a human centipede effect. Elon's reply was sadly quite true. It takes a lot of work to purge LLMs from the Internet training data. Grok 2, which comes out in August, will be a giant improvement in this regard. There's also a rumor going around that Apple might partner with Google Gemini at WWDC, they announced that they were partnering up with OpenAI to power some of its chat features. It's starting to sound like Gemini will also be an option if you don't want to use OpenAI's tech. Now, this article does say Apple could announce a Google Gemini deal this fall, so this is all still just rumor and speculation. WhatsApp appears to be getting a new feature based on some leaked screenshots. It looks to me very much like what Apple showed off in their WWDC keynote, where you can upload an image of yourself and it will generate sort of cartoon or alternate versions that look like you. And this appears to be a feature that they're going to roll out inside of WhatsApp. And I imagine if it's in WhatsApp, it's probably also going to be an Instagram and Messenger and all the other suite of tools that Meta has out there. And since I mentioned Meta, they're getting some competition with their Meta Ray Bans, which personally I love. They're my daily wearers for sunglasses. But there is a new company that is trying to make something very similar that looks very similar with little cameras on the side. And it will have an LLM. In fact, it's going to use ChatGPT4.0 for their LLM inside of it, which, if I'm being quite honest, is still better than what you get with llama 3, which is what's currently in the glasses that Meta has. And finally, this was super cool. I wanted to show this off. It's called Open Television. It lets you immersively operate a robot, Even if you're 3,000 miles away, just like the movie Avatar. So we can see somebody over here wearing an Apple Vision Pro, moving their hands around. They're over at MIT in Boston, and the robot itself is over here in ucsd. And what he's doing with his hands, the robot is actually doing on the other side of the country. I thought that was pretty cool. I want to dive more into this tech. I'm definitely going to do more videos on robotics and things like that in the future. Heck, I'm in San Diego, maybe I can get a demo of this in person over at ucsd. But I just thought this tech was really cool. Wanted to share it as my final little exciting, fun thing for this week. Again, fourth of July week. So not a ton of huge major announcements, but still plenty to talk about. Hopefully you feel more looped in on what's going on in the AI world and found some of this stuff cool, interesting, fascinating, disappointing, scary, whatever. All those emotions are valid. All I know is I love researching it, I love turning around and pointing it out to you. And I'm so grateful that you check out these videos and allow me to do this because I'm having so much fun and I really, really appreciate you tuning in. Don't forget to check out Future Tools IO where I share all of the latest AI news and the coolest AI tools that I come across. Join the free newsletter if you haven't already. It's going to be focused on AI news but also more emerging tech, not just AI. There's a lot of AI newsletters out there, so I'm going to start expanding it into a little bit more of other emerging tech outside of AI as well. You're going to want to see this really cool stuff coming out of that. Check it out at FutureTools IO. And if you like videos like this, you want a weekly breakdown of the AI news, you want AI tutorials, you want to stay in the loop with emerging tech and where the future is headed, like this video and subscribe to this channel and I'll make sure more stuff like this shows up in your YouTube feed. Thank you so much for tuning in. Once again, I really, really appreciate you. Hopefully I'll see you in the next video. Bye bye.