AI Weekly: OpenAI's Advanced Voice Mode and Microsoft's AI Competition

OpenAI rolled out new advanced voice feature, endorsed Senate bills, and faced copyright lawsuits.

### The AI Weekly Roundup: OpenAI's Advanced Voice, Google's Gemini 1.5 Pro, Meta's Segment Anything Model, and More
* 🗣️  OpenAI has started rolling out its Advanced Voice feature to select users, showcasing capabilities like **imitating a frog's voice** or **counting rapidly**.
* 🤖  OpenAI also introduced **GPT-40 Long Output**, offering **up to 64,000 tokens per request** for Alpha participants.
* 🔍  Microsoft, despite owning 49% of OpenAI, now considers it a competitor in AI and search, highlighting the **evolving dynamics in the industry**.
* 🏛️  OpenAI endorsed several Senate bills related to AI regulation, positioning itself as a **key player in shaping future AI policies**.
* 🇺🇸  The White House stated that there is **no current need to restrict open-source AI models**, emphasizing the **importance of maintaining progress in AI development**.
* 🔮  Google released Gemini 1.5 Pro, which currently leads the **lm.org model leaderboard**, showcasing its **AI language model advancements**.
* 📷  Meta introduced the **Segment Anything Model 2 (Sam 2)**, allowing users to easily **segment objects in images and videos**, with potential applications in **video editing**.
* 🎨  NVIDIA and Shutterstock collaborated on Edify 3D, a new AI tool for creating 3D models from text or images, **expanding AI's impact on the 3D modeling field**.

### The Latest in AI Tools and Technologies: A Roundup
* 🖼️ ** Rapid 3D Asset Generation**: Stable Fast 3D's new model allows for **quick 3D object generation from single images**, outperforming Nvidia's Suter model in **speed**.
* 🎨 ** Black Forest Labs' Flux One**: A new text-to-image model is **openly available for use**, with multiple platforms offering **access to it**.
* 📹 ** Runway has expanded its AI capabilities with a new **image-to-video model**, allowing users to **animate still images**.
* 🤖 ** AI Content Creation Tools**: Tools like RenderNet's "Narrator" and Captions' AI Avatar are enabling **easy creation of AI-driven content**, including **AI influencers and digital twins**.
* 🎥 ** AI in Video Production**: Vimeo introduces **automatic video translation using AI**, while the video game industry faces potential strikes over **AI usage concerns**.
* 🌮 ** AI in Everyday Life**: Taco Bell is implementing **AI in its drive-throughs**, while a new AI toothbrush claims to **improve dental hygiene**.
* 🏅 ** AI in the Olympics**: AI is being used in various aspects of the Olympics, from **advertising to sports analysis**.
* 🤖 ** AI Drama in the Tech World**: The launch of the "Friend" AI device sparks controversy, with **accusations of domain price inflation and competing products**.

this has been one of the busiest weeks in the world of AI that I've seen in a long time. I mean, just look at the tabs that I have across the top of my screen. Here, I have a lot to cover in this week's video, so please bear with me if it's a little bit longer than normal. There was a lot of interesting things that happened, and I'm going to do my best to move through them as quickly as possible, but I also kind of want to show some of it off, so let's just see how this goes. Let's start with the thing that seems to be the most viral if you're hanging out over on X, which is the fact that open AI is starting to roll out that new advanced voice feature, you know, the one that they showed off and it sounded like Scarlett Johansson, and then they delayed it for a while. Well, to a select few, they're starting to roll it out. Unfortunately, I'm not one of those select few, so I can't demo it for you myself, but if we

look at open ai's preview here. If you do have it, you should see this option to try advanced voice mode down in the bottom where you would normally type your chat. If you have that, you'll see a screen that looks like this. You're invited to try advanced voice mode, and it will say Advanced Alpha at the top. If you do see this in your app, congratulations, you're one of the lucky ones, and those who have gotten access to it have been showing off some pretty interesting things that it's capable of, like this example from the burgara YouTube channel. Hey, if a frog sang 'happy birthday,' what would it sound like? Hm, I guess it would be something like 'ribbit hoppy bark 32 you Ribbit.' Quite the amphibious serenade. Here's another example here from Cristiano Giardina, where he asked it to speak like a pilot. Ladies and gentlemen, this is your captain speaking. Now as we cruise at 35,000 ft, I have a little story for you from the flight deck. Many.

years ago. Okay, can you actually sound like an airline pilot, like, uh, as if you were talking through the intercom. Ladies and gentlemen, this is your captain speaking. Now, as we maintain our cruising altitude, I thought I'd share a little story from the flight deck. Now, you also notice in that video that he was able to interrupt it while it was speaking. That's a feature that is part of the advanced voice mode as well. If you just have the standard chat GPT app, you'll notice you can't interrupt it like that. And finally, here's a demo of Cristiano trying to get it to count as fast as possible. 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50.

it's crazy 'cause you actually hear it try to take a breath so it's counting really fast, and it's a machine. It doesn't need to take a breath, but the training that's in there knows that if people are counting that high that fast, it stops to take a breath, and it actually does that inside of the app, which is really fascinating. But again, in typical OpenAI fashion, they announce it, they show it off. There's people out there who got access, but most of us still can't use this yet. Speaking of new features that OpenAI rolled out that we can't use yet, they also rolled out GPT-40 long output, which is an experimental version of GPT-40 that has a maximum of 64,000 output tokens per request. So when you ask it a question, it can give you a much longer output. Now, this is something that is available in the API to Alpha participants, and most people don't have access to this yet, but it is designed to respond with extremely long outputs from your questions. While we're on the topic

of open AI. This is something I found interesting. Microsoft is now claiming open AI is a competitor in AI and search. Now, this is interesting because Microsoft has famously invested $1.13 billion into open AI, and they own 49% of the company. But this week, they released their financial reports, and under the competition category for AI offerings, they list companies including Anthropic, open AI, Meta, and other open-source offerings as competitors. And not only is open AI 49% owned by Microsoft, but Microsoft also has partnership deals with Meta as well. Last week, I talked about how open AI announced search GPT, and in this financial report here, Microsoft actually says, Our search and news advertising business competes with Google, open AI, and a wide array of websites. So now Microsoft is even seeing open AI as a competitor to their search business, which is very weird to see a company that's a 49% owner in another company claiming that that company is now a competitor. It's just a really interesting dynamic. Open AI this week also endorsed a few Senate bills. One of

them being the future of AI Innovation Act, which would formally authorize the United States AI Safety Institute as a federal body that sets standards and guidelines for AI models. They also endorsed the NSF AI Education Act and the Create AI Act, which provide Federal scholarships for AI research and establish AI educational resources within colleges and K through 12. Now, as to why OpenAI is endorsing various AI regulation, OpenAI is a likely candidate to face regulatory scrutiny moving forward. So, these endorsements help OpenAI bank goodwill with Federal lawmakers. And even more than that, endorsements like these help OpenAI get a seat at the table on future conversations about AI regulation. So, likely this new regulatory board, the United States AI Safety Institute, members of OpenAI, probably Sam himself, want to be a part of that Safety Institute for that sort of regulatory capture and to steer the direction of the regulation. Also, remember a few weeks ago OpenAI brought on a new board member who was an ex-member of the NSA. So, it seems to me that OpenAI and the US government are

just getting closer and closer with each other. Open AI this week also pledged to give the US AI Safety Institute Early Access to its next model. This article says here that this appears to be intended to counter the narrative that Open AI has deprioritized work on AI safety in the pursuit of more capable, powerful generative AI technologies. Now I'm not an expert. I literally don't even know anybody that works at Open AI. I don't talk to anybody at Open AI, but to me, it just sort of feels like Open AI and the government are trying to get closer and closer and closer together so that maybe Open AI can steer the regulatory narrative a little bit. But that's just an outside perspective from somebody who's kind of trying to keep his finger on the pulse. That's what it looks like to me. And while we're on the topic of regulation, this week the White House said that there is no need to restrict open-source artificial intelligence, at least for now. So right now the US government is basically

saying, We recognize the importance of open systems, the report says. Current evidence is not sufficient to warrant restrictions on AI models with widely available weights. Basically, the government is a little bit concerned about AI and where it's all headed, but they don't want to slow progress. Putting huge restrictions on open-source models could slow progress and allow other countries to essentially catch up or surpass the US in AI advancement, which they don't want to happen. Moving on to Anthropic News, if you're in Brazil, good news for you. You now have access to Cloud. This week, today's video is sponsored by Qualcomm. I recently had the opportunity to visit Qualcomm's headquarters here in San Diego. The work that they're doing over there is making it easier than ever for companies to run AI on device. Qualcomm's innovation is staggering. They hold over 160,000 patents and average about 10 new patents issued every day. I witnessed firsthand how they're turning sci-fi concepts into reality. They had all these little demo

Booth set up, and I saw some truly impressive technology on display. For example, they have these AR glasses that translated any language in real time and showed it on the display inside of your glasses. They showed off technology that can watch your facial expressions and use various facial expressions as inputs for your computer, so that you can navigate what you're doing on the computer just by doing various facial movements. And this really cool AI-powered DJ software that can isolate individual tracks in the music, so that you can overlay them with other tracks, remix them using AI. They showed off a reliable deep fake detector that can watch videos and tell you if what you're watching was a deep fake video or a real video. They had this biometric Palm scanner that was like straight out of Star Trek, where they scanned their hand and it opened a door for you. And probably the coolest thing I saw was this Snapdragon-powered futuristic car that had AI built into it, and it can chat with you. And you're able to generate AI images directly inside of

the car on the display. It was really fun to get to experience some of this cool futuristic Tech that qualcomm's technology has been enabling. What's exciting is that qualcom is making this technology accessible through their AI Hub. AIUB is a platform you can use to optimize models, and it also comes with a collection of already optimized models. As a developer, you can also use the Hub to test performance on real devices on the cloud before deploying them to your device. If you're interested in The Cutting Edge of AI and mobile tech, or if you're developing looking to leverage these tools, check out qualcomm's AI Hub. I'll make sure it's linked up in the description. And thank you so much to Qualcomm for inviting me out to this event and for sponsoring this video. Google has been making some big waves in the AI world this week as well. They released a new version of Gemini 1.5 Pro. They're calling it version 0801, which is August 1st, and it is available to use.

right now inside of Google's AI Studio, you go to AI Studio.g google.com. Over on the right side, we can see a drop-down for model, and you will see Gemini 1.5 Pro experimental 0801. That is the model you want, and you can chat with it here just like you would chat with Chat GPT. Let's pull an image in here and say, What am I looking at? In 5.2 seconds, here you can see that we got a response that describes what we see in the image. Now I am probably the worst person in the world at comparing large language models. I usually turn to someone like Matthew Burman for that. But this new model that Google just released just topped the leaderboard over on lm.org. If we go to chat.lm.org here, click on the leaderboard, we can see that the number one model currently is Gemini 1.5 Pro. It's even outperforming GPT-40, GPT-40 Mini, CLAE 3.5, Sonet, and everything else.

if you're not familiar with how this works, this is a user rating system. So people can go to this Arena Battle, type a prompt here. So if I asked a question like how many breeds of cats are there, and I hit enter, you can see that it's going to give me two responses up here. But it doesn't tell me which model each response is from. It just says Model A or Model B. And then I choose, is A better, B better, is it a tie, or do both suck? Model A gave me a little bit more information, so let's go ahead and say A is better. And then it tells me exactly which model was which. You can see here that I voted Gemini 1.5 Pro as better than what GPT-40 Mini just gave me. And that's where this leaderboard comes from. People are giving it prompts, they then decide whether prompt A or prompt B gave them a more desirable outcome, and then the total winners are summed up here on this leaderboard. So based on just user feedback, Gemini 1.5 Pro is currently the

best Google also released a new smaller version of their Gemma 2 model this week. In the past, they released a 27 billion parameter model and a 9 billion parameter model, and this new model is a 2 billion parameter model, a smaller model built for faster performance and efficiency, most likely for things like mobile devices. They also released Shield Gemma, which is a suite of safety content classifier models that will help filter the input and outputs of AI models. So, it's an open-source tool that helps you essentially put guardrails on your Gemma 2 AI products. But for me, the most interesting news is the Gemma 2 2B model because we can see here that this 2 billion parameter model actually outperforms Mix dral 8X 7B GPT 3.5 turbo and llama 2 70 billion. So, a 2 billion parameter model in Chat Bina, which we just looked at a minute ago, is outperforming much larger models. But that's not the only news Google dropped.

this week, they also added some new Chrome AI features, including new Google Lens inside of Chrome desktop. Top where you can select an area of an image, and it will basically search for that. They actually highlighted this suitcase in this image here and just segmented out that one little part of the image, and it was able to go and find similar suitcases on the internet where you can go and buy them. Another one where they highlighted just a plant here in this image, and it was able to go and figure out what type of plant it was. And this is all built into the newest version of Chrome with this little lens button here. They also added this new compare feature where you can compare products that you have open in different tabs. So in this example here, you can see they've got three different speakers open in three tabs, and then in this tab, it's actually comparing the three speakers that they had open in the other tabs. So in the same way, you might look at a site like Amazon, and Amazon might show you the competitors in a graph towards the

Bottom now you can do this on third-party sites. So these are three different websites, not Amazon, not Google shopping, and it will actually do the comparison for you right inside of a Google window. They also added the feature to speak with natural language to your search history so you can ask things like what was the ice cream shop I looked at last week, and it will look through your search history inside of Chrome. And with that natural language prompt, help you find the information that you're looking for from your history. If you're excited in waiting for Apple's AI features, their Apple intelligence features, well, they're going to be coming a little bit later than initially anticipated. It does not appear that they're going to be in the newest versions of iOS that are going to be rolling out soon. But according to people with knowledge of the matter, the update should be coming by October. This week, Meta killed one of the features that it announced at last year's Meta Connect. If you remember, they showed off these AI chatbots that looked like famous people.

but they weren't the actual chat bots of those famous people. They just used their face, but were trained on different information. Like, for example, Snoop Dogg was the dungeon master, and Mr. Beast was Mr. Bro. Well, nobody really liked those, so they got rid of them. But they replaced it with something else, which is now anybody can create their own custom AI. They rolled out this AI studio, and I love to see it 'cause one of my actual buddies here, Don Allen Stevenson, is one of the people who got it early. And we see his example here. But this new feature allows anyone to create AI characters based on their interests. So, if you go to ai.mea.com AI Studio, we can actually create our own AI here. And we have options like AI pet, private tutor, fellow fan, imaginative artist, sounding board, creative designer, personal stylist, all sorts of random stuff in here. But let's create a custom AI character, and let's do someone who teaches people about AI. It generates a character image with AI, gives it a name and tagline. Then we can create the AI.

and from within here, we can now basically design what we want this AI to do. To me, this looks very similar to GPT-3 inside of chat GPT, but we can give it a whole bunch of custom instructions. We can give it example dialogue, we can change the introductions and capabilities, and really sort of dial it into what we want it to be. Right now, it's interesting. It doesn't seem super valuable to me because I can't, like, pull in large documents or like transcripts of me on podcast or transcripts from my YouTube videos and allow people to chat with a sort of AI avatar version of me. But that's where I kind of see them taking this in the future. That's what I imagine they're trying to get to. But right now, it seems like kind of a novelty. I don't really see how a lot of people will be excited about this yet, but I also kind of see where they're going with this. The more impressive thing that Meta rolled out this week is their new Sam 2 model or segment.

anything model. This is a model where you can segment out certain sections of an image or a video. Now, they already had segment anything model. This new version two just seems to be so much better at, you know, following along to the actual thing that's masked off in the image. Here's an example of a skateboarder where they were able to segment out the skateboarder. They erased them. They added these little like arrays coming out of the skateboarder. Some really cool stuff that it will help with, like no longer needing to do rotoscoping when you're doing video editing. Here's more examples where this person has three balls, and it keeps track of which ball is which. Notice in this video clip here, you got this kid who disappears behind the tree, and when the kid pops back out, the mask still knows to follow the kid around. The older models, as soon as the kid went behind the tree, it would just lose him. And this is actually something that you can play with right now if you want if you head

on over to Sam 2. metad demolab decom. You can actually try it now. By default, you can see it gives us this video of this guy here kicking a soccer ball around. And let's say I want to mask off his shoes. I can just click the one shoe, click another shoe. I'll click track objects, and we can see that throughout the whole video, it continuously tracks where the shoes are on his feet. Let's go ahead and click Start over and change the video. We have a few other video options here, or we can upload our own. Here's a video I took on a recent desert trip of somebody jumping with a motorcycle. Let's see if I can segment them. I'll go ahead and select the motorcycle here, select the person on the motorcycle, click track objects, and now we can see that it actually tracks the person on the motorcycle all the way until they sort of disappear off the screen. And then if I go ahead and click them again right here and then track objects again, it should do a better job.

throughout the rest of the video. Now, now when I watch the whole video back, we can see our motorcycle rider is tracked all the way through the video. I can't wait until they start implementing this into tools like DaVinci Resolve or Adobe Premiere because while they have some like AI rotoscoping built in, this is so much better than what's currently built into tools like that. This week, Perplexity announced the Perplexity Publishers program, which is a program where if they're using your content as a source of news that they give back to people, they'd like to share some of that revenue. But it's not like everybody who has content on the internet is going to get some revenue if their content is shared via Perplexity. It appears that it's only with specific partners, and right now the first batch of partners are Time, The Spiegel, Fortune, Entrepreneur, The Texas Tribune, and WordPress.com. I'm hoping that in the future they'll incentivize just normal bloggers and content creators and people like that to license their content into

perplexity, as well, but as of right now, this partner program is really only for larger Publishers. It seems and news that's actually huge to me personally, but also pretty big news in general. Leonardo AI just got acquired by Canva this week. Now, it's big news to me personally because I've been an adviser for Canva for a little over a year now. It's been one of my favorite AI tools to use, and now they're part of the Canva family. Now, Leonardo is still going to operate individually. There will still be a Leonardo app. They're still going to continue to update and implement new features into it and do really, really cool stuff with it. But also inside of Canva, eventually, we're going to have access to the Leonardo models. Right now, Canva's AI image generator is kind of not great, but once it actually has Leonardo's model like their proprietary Phoenix model built in, we're going to be able to generate really, really good images directly inside of Canva without having to jump around between multiple

tools. It genuinely will be really helpful to me, and I'm really excited about it, but it does also sort of benefit me from an equity standpoint, so not going to downplay that either. While we're on the topic of AI image generators, Mid Journey got an update this week with version 6.1. Version 6.1 greatly improves image quality, coherence, text, and comes with a brand new upscaling and personalization model. Let's head on over to MidJourney.com here. Click on our little settings, and by default, we're set at model 6.1. Here, if I do a penguin holding a sign that says Mreow, I always love testing text that's not actual words to see if it can figure it out because sometimes these models, if you don't use real words, it actually doesn't copy what's in the quotation mark. So let's see how this one does. Not horrible. It seemed to struggle on the m a little bit on all of them, and one of them had actually got right right out of

the gate this top right one's pretty good. All the rest of these are decent, but that M's got a little extra hump on it. But this is so much better at text than it was even a couple weeks ago, so I can't complain too much. And from what I've seen on X, there have been some really impressive images coming out of this new MID Journey model. I mean, look at the realism on some of these images. If you just saw this on Instagram or something, there's very little to see in this that would hint that this is AI. I mean, here's another one. This one, uh, you could kind of tell there's something a little more uncanny about the face here. And here's another really good one. It's making impressive images, but it's just so cool to me to see how far this stuff has come just in a couple years. Now when it comes to AI to 3D models, we've got some new things in that world as well, including a new model that was done in collaboration between Nvidia and shutter.

stock called edify 3D. You head on over to build. nvidia.com. You click over in the Discover Tab and scroll down a little bit. You'll see this edify 3D where you can click into it. It's got text to 3D and image to 3D. But let's just go ahead and try text to 3D. And let's just do a gry wolf howling at the moon and generate preview. And we can see it generated two different previews of a wolf from all sorts of different angles. And I can pick which one I think looks better and click generate 3D. And it did take a couple minutes but then we eventually got our 3D model. Our wolf apparently is blind and has no eyes but not too bad. We can see that it generated this model from all directions. I just wish it gave it eyes but there is another new 3D model generator that claims to be super duper fast stability AI introduced stable Fast 3D which is

rapid 3D asset generation from single images. So, this one, you give it an image and it spits out a 3D object based on that image. This model is available via the API and on Stable Assistant, which I believe you've got to pay to use. Yeah, they have a free trial, but you do have to pay to use this Stable Assistant tool. However, they also put it on Hugging Face, and if we go to the Stable Fast 3D on Hugging Face, we can see that there is a space available. We click into here, we can drag and drop an image right here. Let's just go ahead and use one of the sort of sample images they gave us. Click run, and in less than 1 second, it generated a 3D image. So, quite a bit faster than the Nvidia SL Suter stock version. Let's clear this here. I'll pull in this image of a car, and now I've got just the car without the gray background. And let's run this, and yep, once again, it

only took about 1 second, but as we can see, yeah, not amazing. If we look at all the examples, they're all basically the thing straight on. Seems like if it's at a slight angle, it might mess it up a little bit, okay, a lot bit, but it is really dang fast. All right, moving back to AI image generation for a second. There's a new company in town called Black Forest Labs. And Black Forest Labs is some of the people that helped initially build stable diffusion. If we look here, the Black Forest team, they're a team of distinguished AI researchers and engineers. Our innovations include creating VQ-GAN and latent diffusion, the stable diffusion models for image and video generation, stable diffusion XL, stable video diffusion, rectified flow transformers. They've been some of the researchers and engineers behind some of the biggest advancements in visual generative AI. And with their company launch with Black Forest Labs, they also released a new model called Flux One, which is a new text-to-image model. And this model is openly available for anyone to use. There's a few ways to do

it actually one of them would be using this site called glyph. You can see in this expost here flux is a new open source text image model from Black Forest labs and it's now available to build on glyph. So if we pop on into glyph doapp here this is basically a workflow Builder to build little AI apps but we want we can come up here to build start with building glyph blocks. We'll start with a text input. I'll just say enter your prompt add another block here. We'll select image generator for describe the image. We'll use the input from right here if we want to try to optimize The Prompt we can do that as well by adding some extra inputs like beautiful highquality 4K whatever you want to input that typically gets you better outputs under image generation model we have a whole bunch of options here let's go ahead and select flux Pro. Let's do a 16x9 aspect ratio and we'll leave everything the same under Advanced steps and we will close out of this now we have our glyph available with our.

workflow, and I can just enter any prompt I want. So let's do a wolf howling at the moon, and then run this glyph. And just like that, I got my wolf howling at the moon, and it actually looks pretty dang good. I don't have any complaints about that image at all, other than I thought I was doing 16x9. Oh, but it's reversed. I actually wanted the landscape 16x9. I just, it's the different order than I'm normally used to. I feel like portrait should be 9 by 16, and Landscape should be 16 by 9. I just overlooked that. Another way that you can try out flux is it is also available over on Hugging Face right now. So if we go over to huggingface.co/spaces/BlackforestLabs, etc., etc., we also can run it right here, although we do have a little bit less options to run it with. But let's go ahead and do a monkey on roller skates and click run. And we get a monkey on roller skates using the new

flux model a little funky missing some wheels but not bad overall. It'll be interesting to see if this new Black Forest Labs becomes what Stability AI used to be back when they had Aod Mustak on their team, and they seem to be innovating and rolling out cool new stuff all the time. Maybe Black Forest Labs is going to fill that gap. But moving on to AI video, it was also a big week in AI video because Runway, who released Gen 3 Alpha a few weeks ago, now rolled out an image-to-video model. Before, they only rolled out the text-to-video. Now we have image-to-video, and it's actually really solid. We can see some examples of water falling on this guy here, some plants falling on this. It looks like a Suzuki Samurai. This statue here getting covered in paint, all sorts of really cool videos that started as images. And if we just head on over to Runway ml.com, click on Get Started under the Gen 3 Alpha, here we actually have a giant box where we

can now upload an image. Let's pull in this image that I created with these hands with paint on them. It does seem to want like a 16x9 or a landscape resolution, so I need to crop it here. And then we can add a prompt. Let's say hands with colorful paint start clapping. Let's go ahead and generate and see what we get out of it. I have no clue if this is going to look good or not. And of course, as soon as I go to record this video, they are experiencing high demand. Gen 3 Alpha Generations are unavailable. Please wait for your generations to complete or try again later. So I don't know, maybe they're happening in the background. I don't know what's going on right now. I did mess around with it earlier though. Let's go ahead and open assets. I was uploading some of my YouTube thumbnails. Here's one where I was wearing VR goggles and it actually animated this video of me in VR goggles. Here's another thumbnail that I uploaded that had colorful feathers with me looking at

the camera so it can make some pretty dang cool videos. I guess when the demand isn't too high for it to be able to make cool videos. Runway also announced this week their new faster, cheaper AI video model gen 3 Alpha turbo. We can see from this demo that they put out that when you're actually using gen 3 Alpha now, it actually runs a lot faster. So when you enter your prompt, you actually get your video output much faster than you used to. In fact, it shows that it only took 11 seconds to generate this using gen 3 Alpha turbo image to video. And over on the left, it's still generating. It looks like it's only about 25%. So yeah, a lot faster. Lately, we've also been seeing a lot of apps that are kind of like deep fake Creator apps. I don't know what else to column like this new one from render net called narrator where you can upload a video at a script and then it will seamlessly lip sync the character to your word. So here's the

example on this one. Hey y'all, you ain't going to believe what just went down. My cat hacked my computer. I mean, I don't think it's going to really fool anybody, but it looks pretty cool and it could be fun for content creation. There's a whole bunch of other examples. This is one I haven't played with yet myself. There's a woman sitting in like a coffee shop. Here, oh, this morning I got into the craziest fight with Captain America over his shield. I mean, the lip syncing isn't perfect. You can definitely tell, but you can check this out over at Render That. And what I think is even cooler about Render That is not necessarily like the lip syncing and the audio. It's that you can create a super consistent character that you can use over and over again. So anybody that is interested in creating like a digital influencer, you can essentially prompt this character into existence, right? Build your unique character, design them how you want them to look so they're consistent every time you use this character. You can put that.

same character into various poses. You can change their clothes, and then once you have that character that doesn't exist in the real world, you can then do the voice-over and lip-syncing and stuff like that. I think the technology is really cool. I am a little bit scared of a flood of junk content and all of the YouTube videos about how you could make a million dollars a year by creating AI fake influencers. Like I really kind of hate that concept, but the tools that are available to do that kind of thing are getting more and more readily available. Like RenderNet AI. While we're on the topic, Captions also released a similar feature. Here's a little demo video that they put out where they created an AI Avatar. You can see the AI-generated version on the left, the real version on the right. Kind of the same concept. Can you tell which one of us is AI-generated? That's my twin. I generated her in a minute, 100% Captions. I guess the big differentiator is that RenderNet, that we just looked at, is for creating sort

of non-existent AI influencers. Where this one is to create a sort of digital twin of yourself, which seems slightly more valuable. I know some people that have been very very successful by making commentary videos like I make on YouTube, but they're actually using a sort of digital twin version of themselves. It still looks like them, they still look like they're talking, but they actually let AI create the video motion of themselves, which could help people make a lot more content at scale, but also a lot more low effort content. So once again, I put this in the box of I'm excited by the tech, but sort of not too excited about the implications of it, I guess. And since we're on the topic of AI video, Vimeo, which used to kind of be a YouTube competitor, but now it just seems to be more of like a video hosting platform, is rolling out a new feature where it will automatically translate your videos using your own voice into any language you want. Now, it doesn't sync the lips, it just translates it.

uses your voice, but if you use Vimeo, it seems like it's a pretty handy feature so that you can localize your videos to wherever you're publishing them. Thank you for purchasing a membership about Hindi exclus or Portuguese this week. Also, responded to the lawsuits that have been filed against them claiming that they were training on copyrighted information. I'm not going to read this whole thing here. I will link it up below if you want to read it, but the gist of it is they say that they used publicly available data all over the Internet. And yes, when you're trying to pull in just tons and tons of publicly available data, unfortunately, sometimes that data includes copyrighted information. They weren't specifically going to let's say Spotify and Apple Music and scraping music from there, but if they scraped from a website that was publicly available that maybe had that music embedded on it, it could have ended up in the training data, but that wasn't their intention. They also go on to claim that Sunno learns just like a human learns so.

if a human wanted to learn how to play a specific type of music, they might listen to a whole bunch of people playing that style of music and learn from all of that input that they got. And Sunno essentially learns in that same way. They also go on to say that they were surprised by the lawsuit because they were actually working with a lot of people in the recording industry who were excited about the tech, so it sort of felt like it came out of the blue. They also talk about how they've put guardrails in place so that you can't actually generate music that sounds like specific artists. Like you can't enter a band's name in the prompt to get music that sounds like that, and that the whole tool was designed to create original pieces that don't actually sound like any specific artists. Again, they make some really good points. It is a long post and it will be linked up in the description if you want to hear their rebuttal to these lawsuits. This has been fascinating to watch if you're

on xit all. You've probably seen this already, but there's this new AI device called friend. And I'm not going to play the whole video. I don't know the copyright status of the music on it and things like that, but basically this friend Gadget is this little necklace that you wear that is listening to you all day long. And as it listens to you, it actually will send you text messages like, well, at least we're outside. So she named her friend Amy, and Amy realized they're outside and texted her, well, at least we're outside. In another example, this dude is playing a video game with his real life friend while wearing his necklace friend, and he's getting text messages like, you're getting thrashed. It's embarrassing. Here's somebody watching a show on their phone, and their AI friend Emily says, this show is completely underrated. It also notices that she's eating here and says, how's the Falafel? The girl actually responds by just speaking out loud. She drops some of the sauce from the Falafel on her.

friend and the friend goes yum. You can pre-order it for 99 bucks. Personally, I actually pre-ordered one. I'm a sucker, but not because I actually think it's a good idea. I just want to play around with it and make videos about it and make content around it. Personally, I don't get it. Like why do I want something that's just going to listen to me all day and then send me pointless text messages like you're getting thrashed in the video game you're playing? I guess I'm not the demographic. I don't understand it. I don't get why people would want this, but I did order one. So when I get it, I'll give more informed thoughts on it, I guess. But this has been more of a story than just the announcement of this friend product. In fact, there was another product that was a necklace that was also called friend, and it was launched by Nick shenko. Here, and when this new product called friend came out, which was launched by Avi shiffman, well, Nick here made a rap.

rebuttal. So, check this out. I'll let you know in advance, there's a bit of a cringe factor to it, but let's take a look. Yo, I'm Nick, and for those who don't know, I build PR, rammed your TAP, jacked my style, build nothing real. It's been a while, invest their cash, you've been quick. I'm grinding hard to it, my Peak. Don't be game, where is the proof, all no roof. I'm the founder, hardware in hand, you're just a shadow, can't understand. I your jokes, SP a meal on a name, can she a thing, just Twitter Fame. FR and yours running my wave, open source fam, diing Your Grave. Take a seat, I learned the grind, you're wasting cash, I'm building in time. Respect the game, respect the source, I'm the real friend, the eii force. So what he's referring to is his initial version of friend, which he launched as open source. You can see it here over at basehardware.com, and it's kind of the same concept, right? It's a necklace that you

wear but it actually looks more useful then the new version of friend. Like it's got translation features, event reminders, music, health indicators, emails, task shopping list. It's got a lot more built into it and actually seems a lot more useful while also being open source. Not only was the concept basically lifted from somebody else and the name taken from an existing product, but there's even more to the story. It gets even more interesting. This new friend product, the Avi shiftman version, they recently raised $2.5 million, and of that $2.5 million, they spent almost $1.9 million of it buying friend.com. Oh, but the story gets even more funny. Here we can see that Macy a Gilliam here, well, she posted a tweet that says congrats to whoever sold these idiots this domain for 1.8 million. The only person who's going to make money here. And then they linked off to that same friend video that, uh, we were just kind of skimming through a minute ago. One thing of note that's fairly interesting here is if we look at

Macy's name. She works for Morning Brew. And well, here's the thing about Morning Brew. Remember that 2.5 million I just mentioned? They raised. One of the investors is Morning Brew. So Austin Rief here, who is the CEO and co-founder of Morning Brew, is talking about the viral launch of this friend tool here. Now let's jump back to Macy's thread here. Scroll down a little bit and we can see that Avi Shiffman here points out, Hey, you work for my investor. But Macy here has the ultimate burn, Oh, so you work for him too then. And uh, check out that ratio there. Anyway, it's all just been fun to watch. This is like my version of The Real Housewives. I love watching all these like tech company drama stories unfold. I just find it so much fun and so fascinating to watch. A lot of people are speculating that all of this drama between the two friend product and Macy calling them out on the price, it's all part of the marketing because people like me are now talking about it. In

videos like this. So who knows, maybe it's genius marketing. At the end of the day, I don't know. I'm just watching it all play out with popcorn in hand, thinking it's fascinating and fun. But, uh, yeah, there's some interesting storyline happening there. All right. There's a handful of other news that came out this week, but it's all a little bit smaller news, so I'm going to sort of wrapid fire it here for you. I warned you in the beginning this was a long week with a lot happening. If you remember, recently there was a writer strike and an actor strike where AI was part of the reason they were striking. Well, now video game performers are basically going on strike over artificial intelligence concerns as well. They're worried that game companies could train AI to replicate actors' voices or create digital replicas of their likeness without consent or fair compensation. So they're basically striking to try to get the game companies to not do that. This one's going to be interesting. I think this is

going to play out a little bit different than the actor strike because the tools already exist to create new character voices. You can use something like 11 labs to create really realistic AI voices that aren't attached to any real person. You can just create new ones, same with the characters in the games. I'm pretty ignorant on the topic. I'm sure people in the comments will fill me in on my gaps and thinking here. But to me, it doesn't feel like the same thing as like movie and TV actors because it's so easy to just create video game characters without the necessity of somebody acting that out already. But again, I don't know what I'm talking about. Moving on, Taco Bell is rolling out AI in their drive-throughs. They're going to use voice AI technology in hundreds of stores in the US by the end of 2024. Pretty soon, you may be ordering from an AI drive-thru at Taco Bell. However, when companies like Wendy's and McDonald's tried this in the past, it didn't really

work out so well so we'll see if maybe Taco Bell has a different experience. There's a new AI toothbrush, which according to ARS Technica here, claims is coming for your teeth and your data. Apparently, this toothbrush claims to use advanced algorithms and companion apps to help you brush your teeth better. And I really need to make a video on like these are things that don't need AI so just stop. But yeah, we're going to get AI toothbrushes pretty soon and I'll probably get one cuz I'm a sucker for this stuff. And finally, AI has been everywhere in the Olympics this year. If you've been paying attention to the Olympics, you've probably seen a bunch of AI ads from companies like Google. However, the advertising isn't the only place AI is being used. It's actually being used in a lot of the sports as well. We can see here how AI is actually used to identify things on the playing field, to analyze movements of athletes, and things like that. Here's another area.

where AI is being used. Where you can actually see the tracking on the ball. So you we can all get a better idea of how the ball is moving. AI is just being used everywhere in the Olympics, and I'd actually like to make another video where I break down all the ways that AI is being used in the Olympics. So I'm not going to go into more depth in this video. I just found that really fascinating how much AI is being leveraged to actually better judge what's going on in a lot of these Sports. Anyway, that's what I got for you this week. I've been playing a lot of catchup after vacation. I was in Colorado the last couple weeks, so I haven't been putting out as many videos as I've wanted to. But I'm about to pick up the pace again now that I'm back home. I'm in a stable environment. I'm excited. There's a lot of cool stuff that's happening: the AI and the Olympics, all these new AI video tools, new large language models to

play with. There's so much happening that I want to play with and make more tutorials and demo videos. I've had a little bit of a like mental barrier around making videos lately 'cause I've gotten so addicted to looking at the numbers on YouTube and trying to make more videos like the videos that got big numbers, but I have shifted my thinking and I'm just going to go back to my old school style of videos where whenever I find something that excites me, I'm going to make a video about it and put it online. If it excites you, you'll probably watch it. If it doesn't, it won't get views, but I don't care anymore. I'm just going to make videos around the cool AI that excites me, make tutorials on the stuff that I'm learning. I'm going to go back to the old school style of making the videos that I used to make, including the Friday news videos that I've always made for a couple years now, and some of the videos I put out will be for you.

some of them won't, and that's okay. Hopefully, people will enjoy what they want to enjoy and ignore what they don't enjoy. But that's really all I got for you today. If you like videos like this, you want to stay looped in on the latest AI news, you want to see what kind of AI tools and tutorials I think are worth paying attention to right now, make sure you like this video, subscribe to this channel. I will make sure more stuff like this shows up in your YouTube feed. If you haven't already, check out futur tools, where I share all the cool AI tools that I come across. And I also have a newsletter where I tell you about the coolest AI tools directly in your inbox. It's all free. You can find it over at futur tools. Once again, thank you so much for tuning in to this video and nerding out with me today. There was a lot to nerd out about, so this was longer than normal. Apologize for that, but hopefully it was

still fun entertaining informative uh you know whatever you wanted to get out of it hopefully you got that out of this video. I really really appreciate you thank you again to Qualcomm for sponsoring this video. I will see you all in the next one bye-bye.

AI News: The Busiest Week in AI in A Looong Time!

AI, No-Code, Tech, Futurism - I'm a tech nerd and talk about tech nerd stuff

Matt Wolfe

OpenAI is rolling out its new advanced voice feature, allowing select users to try out the impressive capabilities, including interrupting the speech and fast counting.

5 long output, offering a maximum of 64,000 tokens per request for Alpha participants.

Microsoft, despite owning 49% of OpenAI, is now positioning the company as a competitor in AI and search, which creates an interesting dynamic.

Additionally, OpenAI has endorsed several Senate bills, aiming to establish goodwill with Federal lawmakers and secure a role in shaping AI regulation in the future.

Open AI's Collaboration with the US Government and Recent AI Developments

Open AI is providing early access to its next model to the US AI Safety Institute, aiming to address concerns about deprioritizing AI safety.

The US government has stated that there is currently no need to restrict open-source artificial intelligence, in order to avoid slowing down progress and allowing other countries to surpass the US.

Qualcomm is making groundbreaking AI technology accessible through its AI Hub, offering optimized models and the ability to test performance on real devices.

Google's Latest Releases and Meta's New Features

Google has unveiled a smaller 2 billion parameter model, outperforming larger models, along with new Chrome AI features like Google Lens and product comparison.

They have also introduced natural language interaction for search history and announced a delay in Apple's AI features.

On the other hand, Meta has scrapped AI chatbots resembling famous people and launched AI Studio for creating custom AI characters, while also introducing the improved Sam 2 segmenting model for images and videos.

Perplexity's New Publishers Program and Other AI Updates

Perplexity has launched a new Publishers program, initially available to partners like Time and WordPress.

com, aiming to share revenue with content sources.

Canva has acquired Leonardo AI, with plans to integrate its models into Canva's design tool, offering improved AI image generation.

1, bringing significant improvements to image quality, text, and upscaling, while also impressing users with realistic AI-generated images.

Nvidia and Shutterstock have collaborated to create Edify 3D, offering text-to-3D and image-to-3D generation with the ability to preview and customize models.

Rapid 3D Object Generation from Single Images

A new 3D asset generation model, Stable Fast 3D, is available on Stable Assistant and Hugging Face, offering quick 3D object generation from images.

Black Forest Labs has launched a new open-source text-to-image model called Flux One, accessible through Glyph and Hugging Face, potentially filling the gap in visual generative AI.

Runway has rolled out an image-to-video model as part of Gen 3 Alpha, allowing users to create videos from still images with prompts on their platform.

AI video technology is rapidly advancing, with Runway announcing a faster and cheaper AI video model.

RenderNet has introduced a new deep fake creator app called Narrator, allowing users to seamlessly lip-sync characters to a provided script.

Captions has also released a similar feature to create a digital twin of oneself using AI-generated video motion.

Vimeo is rolling out a new feature to automatically translate videos using the creator's voice into any language, while Sunno, an AI music generation tool, has responded to lawsuits claiming they used copyrighted information in their training data, stating that they are designed to create original pieces without replicating specific artists.

The Controversy Surrounding the "Friend" AI Device

A new AI device called "Friend" is designed to listen to users and send them text messages based on their activities, but it has raised concerns about privacy and its actual usefulness.

The "Friend" device sparked a dramatic conflict between two versions of the product, involving a rap rebuttal and a multi-million dollar domain purchase.

The controversy surrounding "Friend" has led to speculation that the drama is a deliberate marketing strategy, while also raising questions about the ethical use of AI technology.

The Growing Influence of AI in Various Industries

AI technology is being implemented in Taco Bell's drive-throughs, following previous attempts by other fast-food chains.

A new AI toothbrush is being developed to improve teeth brushing using advanced algorithms and companion apps.

AI is widely used in the Olympics for advertising, as well as in sports to identify and analyze movements of athletes, showing its increasing presence in the sports industry.