Resources:
Join the discussion: Ask Jordan questions on AI
Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup
Connect with Jordan Wilson: LinkedIn Profile
Try Our Free AI Prompting Course: Register for our free Prime, Prompt, and Polish AI Course!
OpenAI's Controversy over Sora - Text-To-Video Tool
In recent news, OpenAI's new text-to-video offering, Sora, found itself at the heart of a contentious situation, ignited by claims from artists involved in its early access. The artists accused OpenAI of exploiting their labor, arguing their role was reduced to "PR puppets". Furthermore, concerns were raised over Sora's training practices, with particular scrutiny over the potential use of YouTube content. Despite protest and leaks from the artists, OpenAI managed to navigate this tricky situation, emphasizing its liaison with over 100 artists involved.
Anthropic's Model Context Protocol: A Game-Changer for AI Models
To fuel more relevant AI interactions, Anthropic has introduced the Model Context Protocol (MCP). Intended to enhance AI data connectivity, this open-source standard simplifies AI development, offering pre-built MCP servers for popular platforms like Google Drive, Slack, and GitHub. This move comes alongside Anthropic’s customization-friendly writing styles for Claude AI, promising more tailored AI interactions.
Alibaba's Quan Team and Uber's Expansion into AI
In other updates, Alibaba's Quan team revealed an experimental AI model, the QWQ 32B Preview, a behemoth boasting 32.5 billion parameters, designed for complex reasoning tasks and posing a challenge to OpenAI’s reigning models.
Not left behind in AI race, established industry player Uber Technologies has also taken a leap into AI's thriving world, launching a new division, Scaled Solutions, to offer AI development outsourcing services. Using its vast gig economy experience for AI model training and data labeling, the transportation giant is poised to shape discussions on AI's impact on the traditional 9 to 5 job structure.
Audio and Video Advancements in AI
AI's innovation spree has seen a flurry of developments in the audio and video sectors. NVIDIA unveiled Fugato, a groundbreaking generative model for audio synthesis. It's capabilities include blending distinct sounds, producing sounds as unconventional as "saxophones barking".
On the video front, Luma Labs has made significant strides, enhancing realism and motion consistency in its Dream Machine. Meanwhile, LTX Video has open-sourced their tool on Hugging Face Spaces for local usage, with Runway launching new AI video features.
OpenAI Faces Legal Feud while Venturing into Advertising
Spicing up the AI scene is billionaire entrepreneur Elon Musk's preliminary injunction against OpenAI and Microsoft. Alleging anti-competitive practices, Musk's legal team lodges the argument for preserving OpenAI's nonprofit character.
Meanwhile, OpenAI is turning to advertising for revenue as they eye 1 billion users by next year. As the AI giant dives into the advertising pool, the industry holds its breath, curious about the ad models OpenAI will embrace to align with their growth strategy.
Conclusion
AI's rapid development continues to disrupt conventional business models and fuel discourse around its implications. From corporate expansions, legal disputes and Advanced innovation, the AI industry remains a dynamic field with far-reaching impacts across the business landscape.
Topics Covered in This Episode
1. Controversy Over OpenAI Sora
2. Anthropic's Innovations
3. Alibaba's New Experimental Model
4. Uber's AI Expansion
5. NVIDIA's Fugato Model
6. AI Video Advancements
7. OpenAI’s Revenue Strategies
8. Lawsuit Against OpenAI and Microsoft by Elon Musk
Podcast Transcript
Jordan Wilson [00:00:16]:
Did ChatGPT release anything for its second birthday? Why are Elon and OpenAI fighting again? Are we all gonna be working AI jobs at Uber? And is ChatGPT gonna start shoving ads down our faces? So many AI questions this week. We've got your AI answers. What's going on? Y'all. My name is Jordan Wilson and welcome to Everyday AI. If you're maybe wondering some of those questions this week as you're looking to grow your company and grow your career, don't worry. We're gonna be here with some answers. If you're new here, thank you for joining us. We do this every single day.
Jordan Wilson [00:00:59]:
Everyday AI, it's for you. This is your daily livestream podcast and free daily newsletter, helping us all learn and leverage generative AI to grow our companies and our careers. So if you're listening on the podcast, thank you as always. Make sure to check out your show notes. There's a ton in there. But most importantly, maybe is a link to our website at your everyday a i.com. Make sure you go sign up for our free daily newsletter. So, yeah, on Mondays we bring you the AI news that matters.
Jordan Wilson [00:01:28]:
So you don't have to waste hours every single day wondering how AI is going to impact your career or your company. On Mondays you can spend time with us. But we do this Monday through Friday every single day, bringing on some of the best guests in the world and you can ask them questions and get answers live. Right? And you can also go to our website at your everydayai.com. We have more than 400 episodes. You can go listen to old podcast episodes on our site. You can watch videos on our site. You can read.
Jordan Wilson [00:01:58]:
There are there's thousands hours of free generative AI content. So make sure you go to your everyday a i.com. Alright. Enough chitchat. Let's get into the AI news that matters for this week and thank you as always, to our live stream audience for joining us. Michael and Sam and Philip joining on YouTube and, Brian and Jay saying that he's in everyday AI withdrawal from the Thanksgiving, holidays. So, yeah, excited to be back with you guys all and Rolando and Marie and Woozi and Fred and Christopher, everyone else. Alright.
Jordan Wilson [00:02:29]:
Let's get into it y'all. So OpenAI's Sora. We haven't heard about that in a while. What's going on? Well, OpenAI's new text to video AI tool, Sora, is kind of embroiled in a little bit of a controversy following a leak by early access artists, highlighting tensions between tech companies and creative communities. So the artist posted in a hugging face space, as p r puppet Sora. So these artists were granted free access, for testing, and they accused OpenAI of using them as quote unquote PR puppets. That is why the name of the hugging face space there was called PR puppet Sorra. So they accused OpenAI of using them as puppets instead of genuine collaborators, revealing a rift over the company's approach to artist involvement.
Jordan Wilson [00:03:23]:
So, yeah, essentially, late last or or no, early this year. So almost 10 months ago, a group of select, creatives got access to OpenAI, Sora, but there was some processes. Right? You had to sign, you had to sign some some documents. You had to say, hey, we're gonna get these things approved through OpenAI before we release them. You know, OpenAI has said that was to kind of control not necessarily control the quality, but to make sure harmful, content was not getting out through, its new Soarer tool. But these new, artists these artists that just posted this online are not liking the terms that they signed up for. So according to their statement on Hugging Face, around 20 artists claimed they were misled into unpaid labor, arguing OpenAI exploited their creative input for marketing despite the company's $150,000,000,000 valuation. So the group criticized restrictions requiring OpenAI's approval for SOAR generated content publication, suggesting the process was more about promoting the tool than fostering creativity.
Jordan Wilson [00:04:29]:
So in protest, these artists leaked Soarer access. So through a back end API and they posted all of this essentially a Python script on Hugging Face, allowing people to use Sora. It was only up for a couple of hours. I went in and tried it out. So, it did really put some pressure on OpenAI, and OpenAI quickly shut this down. So they responded by suspending Sora's early access 3 hours post leak, defending its voluntary program, and emphasizing the valuable contributions of 100 of artists. So not all participants though agreed with the dissenters as a lot of artists, went online and supported OpenAI stating that this protest from a select few did not reflect the majority view. So the incident adds to existing scrutiny over Sora's training practices with concerns about potential use of YouTube content raising ethical and legal questions.
Jordan Wilson [00:05:30]:
So, we did cover this in our newsletter last week y'all. So, I'm I'm curious though. What what does our livestream audience think? I'm gonna give you my takes on this. So did these artists make a point? Right? So they essentially said, hey, the whole world has wanted this, Sora tool. We're gonna, you know, create a kind of a technical, backdoor and allow people to use Soar. Well, number 1, it took forever. It took if if you were very early, you could get it out in a couple minutes. But for most part, you know, kind of in the middle of this 3 hour wave, it took like an hour.
Jordan Wilson [00:06:04]:
So could people really use it? Not really. And did these artists make a point? I don't know. Right? You're signing up for something. Right? When when you get early access to a program, you are essentially signing away your ability to make money. Right? If and if a company, any AI company says, hey, we're gonna give you early access to these these tools. Here's the rules. Do you want in? Yes or no? I mean, that's what you're signing up for. So I'm I'm not quite sure, what this what this group of kind of rogue artists, was hoping to accomplish.
Jordan Wilson [00:06:39]:
But I think if anything, they just brought more eyeballs to Sora. Right? We hadn't been talking about it a lot recently, and anyone that follows the show, I've I've been saying all along that we're not gonna get access to Sora until after the election. So, you know, I would assume in the coming months that we're gonna be getting access. But a couple of things actually came out from this leak though. So yes, it just kind of was through an API back end. But, we saw a couple, styles. Right? So that's that's interesting. So what that leads me to believe is that we are going to, either have the ability to, to create images with Sora or, you you know, we'll see a DALL E update that essentially uses the Sora engine.
Jordan Wilson [00:07:22]:
Also, it was called turbo. That part is huge. So that indicates, that OpenAI actually had different tiers or different models of Sora. So potentially, turbo version which is the version that, through this hugging face, repo that people kind of got access to via the back end, it was a turbo version. So it was supposed to be faster. So a lot of people are saying, oh, this is the more powerful version and I'm not necessarily sure about that. I think that whenever OpenAI does grant access to the rest of the world to Sora, I don't think we're gonna be getting the most powerful version. If you think back to the turbo, how OpenAI has used this in the past, normally it's actually a dummy down version.
Jordan Wilson [00:08:04]:
It's not the most powerful version and it's just faster. Right? So I don't necessarily think that when OpenAI does actually release this that we're going to get the most powerful version. There's been early reports that we've talked about on the show that they're actually working on a new version, of Sora. So I don't think when it does come out, that we're gonna be getting the most powerful version. We'll probably be getting this, turbo version or a version that sacrifices quality for speed. So I don't know if this worked. I think of anything, like I said, I think it just brought more eyeballs to Swara. And some of the, like, some of the, generations were pretty impressive.
Jordan Wilson [00:08:40]:
Some weren't, but the ones that worked. Right? The thing with these AI video models, just like anything with generative AI, they're a roll of the dice. But the dice, when you're talking about video have way more sides to the die. Right? So, you know, I think that the the floor is much lower. The ceiling is much higher. And for those generations that hit, they looked very impressive. Better than literally anything else out there. So I think if nothing else, you know, this just gave Sora more eyeballs, people talking about how good it was.
Jordan Wilson [00:09:10]:
Could this have been an internal leak? A lot of people are are speculating that, oh, OpenAI did this. I don't know. That doesn't really seem, like a like a, you know, marketing ploy out of their wheelhouse. I could be wrong. So, yeah. Fred is saying artists think differently. Alright. So we'll see.
Jordan Wilson [00:09:31]:
We'll see. Alright. Let's let's get to our piece our our our next pieces of AI news y'all. So Anthropic just been shipping. They have been shipping so much the last couple of weeks when a lot of other big companies, you know, put out previews or they tease or wait list. Anthropic, the maker of Claude is shipping. So Anthropic has launched a new open source standard called the Model Context Protocol or MCP to improve AI models' access to data, potentially leading to more relevant and accurate responses. So MCP allows AI models to connect with various data sources, including business tools and content repositories overcoming the limitations of isolated data systems.
Jordan Wilson [00:10:15]:
So according to Anthropic, MCP enables two way connections between data sources and AI applications, simplifying the development process by offering a standard protocol instead of separate connectors. So, Anthropic also has released pre built MCP servers for platforms like Google Drive, Slack, and GitHub with plans to offer other tool kits tool kits for broader organizational use. So right now, this initiative aims to create a substantial or so sorry. A sustainable architecture for AI systems, maintaining context across different tools and datasets. So despite its potential, the adoption of MCP, faces challenges particularly from competitors like OpenAI, which recently introduced a similar feature in its ChatGPT platform called Work With Apps. However, Work With Apps is a one way street, and MCP from Anthropic is a two way street. Work With Apps is very simple to set up. MCP, not as simple.
Jordan Wilson [00:11:17]:
So you do have to be a little more technical, to be able to work with your kind of internal databases here, with MCP. But it's an open, they open source this, right? So, I do like that piece because, you know, I think the open source community can take this MCP protocol and really run with it and really create, some great ways to connect large language models with, internal data sources in the very near future. So I do like Anthropic's approach here even though you do have to be very technical, right? Just like as an example, we did this live on the show about a month ago showing, off Anthropic's new computer use, where you can essentially talk to Claude and it can literally run a computer like an agent. It's not very good yet, but I do like Anthropic's approach here. They're they're not teasing things, you know, they're just shipping new helpful updates. Speaking of new updates, I don't know how helpful it is, we'll see. But Anthropic has also introduced customizable writing styles for Claude AI. So Anthropic is enhancing its Claude AI chatbot with new customizable writing styles, a move that could transform AI communication by allowing users to tailor responses to their preferences.
Jordan Wilson [00:12:29]:
So maybe we'll have less delving and less revolutionize or less in today's digital realm, content out there. Right? So this development is significant as it aims to make AI interactions more natural, less robotic and more personalized. So there are 3 preset styles available for everyone. So right now, this is available also for free users, although it's much more limited. So the 3 preset styles available are formal for professional and precise responses, concise for short and direct responses, and explanatory for detailed explanation. But the standout feature is the custom style option, which allows users to upload writing samples and descriptions to teach Claude their preferred communication style, and then you can edit those. So over time, users can refine these custom styles, making the AI a better mimic of their personal writing tone and voice. So the introduction of these features raises questions about the role of AI in content creation and the potential blurring of lines between human and AI generated content.
Jordan Wilson [00:13:37]:
So is it any good? Meh. So I mean, I'll say this, and let me explain this. Right? Because I know some people are gonna get angry, but I like to tell people how it is. I use large language models like Claude, like ChatGPT, like Gemini, and then other platforms that leverage large language models, right, like Perplexity. I use these hours hours a day. Right? It depends on the week, but anywhere from 2 to 10 hours. Right? I'm using large language models since the day since not just since they were released, since they before they were released. Right? Since, late 2020 is when our team started using, you know, the early GPT technology.
Jordan Wilson [00:14:20]:
And my background is also as a journalist. So the two things I know most about right now in my life are writing and large language models. Right? Out of the box, Claude is great. It is. It it sounds, less robotic than other large language models. I will say Claude is 1 a, Google is 1 b, and then ChatGPT is number 2 in terms of out of the box, kind of content writing or the ability to sound human. Right? But even with this new style, preset here from open or sorry, from, Anthropic, it's not very steerable. Right? It's kind of hard headed if I'm being honest.
Jordan Wilson [00:15:03]:
So, if your writing style is is kind of vanilla and there's nothing wrong with that, then I think you'll like this. If you have a unique writing style, I don't think you're gonna like this. Right? So as an example, you know, I did try this out. I was gonna put it, you know, do a review. I decided not to, because I didn't think it was that great if I'm being honest. What are the problems? Right? How people write, it's it's it's a style. It's a cadence. And, I think in my many many many many many hundreds of hours of experience, ChatGPT is still better at being steerable.
Jordan Wilson [00:15:36]:
And what that means, being able to, improve and steer, a tone of voice. Where Claude, like I said, is better out of the box, but even with this new style preference, I don't think it's very steerable. What you kind of get in the first iteration, even though you can go conversationally edit it, which I think is nice, It's not that great. And and what people don't see on the back end. Right? So let's say you generally write in short sentences and then you you know, you have multiple paragraph breaks, etcetera. It's just plain text. Right? So you can do all that in your writing, but ultimately what Claude is seeing is plain text and is jumbling it all in giant paragraphs. Right? Which is, not saying like a telltale sign of AI.
Jordan Wilson [00:16:14]:
Right? But it just makes it hard to read. So hey, livestream audience, let me know, are you liking this, this new Claude style? We did put it in our newsletter, obviously. So, let me know what you think. Hey, Justin. Thanks for joining us. Justin said, long time listener. I've been listening to you for a long time. First time seeing you live.
Jordan Wilson [00:16:33]:
Yeah. If you listen to the podcast, come join us live. Right? It's fun. You can network with other people. Ask me questions if you have any. I'll try to do my best. Alright. More AI news.
Jordan Wilson [00:16:45]:
So Alibaba, their Quen team has released a new experimental AI model named qwq32bpreview. Love all these model names that are super easy to say. So, this new model from Alibaba's Quan team is a reasoning model. Alright. So similar to OpenAI's o one. So this new model is notable for its 32,500,000,000 parameters, allowing it to handle prompts up to 32,000 words in length. So it is what they're hoping what the Quen team is hoping to be, a rival to OpenAI's o one model, which was originally, code named, strawberry. Now we all know it is called o one, and it's designed to improve problem solving skills.
Jordan Wilson [00:17:33]:
So, yeah, the biggest difference between traditional large language models and reasoning models is essentially these reasoning models, go through kind of a chain of thought process under the hood, doing kind of the initial prompting work that a human would normally do to drive better results. So these new reasoning models, right, it is kinda like a new tier. So you can't really compare, you know, quote unquote traditional large language models with reasoning models because they're very different, in their functioning. So, the new q w q 32 b preview is one of the few AI models that facts checks itself according to Quinn, reducing common errors, but also sometime taking longer to find solutions. So right now, users can run and download QWQ from the AI development platform, Hugging Face. And despite its advancements, the model has limitations including potential language mixing and circular reasoning patterns as noted by the company itself on its blog. So the model from Quen comes just about a week after r one lite, which is the newest model from China based startup DeepSeek. So, yeah.
Jordan Wilson [00:18:40]:
In the last, like, 10 days, we've gotten 2 reasoning models out of China. 1st, r one lite from DeepSeek, and now qwq32bpreview from Alibaba's Quan team. Mouthful. Are we all gonna be working for Uber? I don't know. Maybe. Alright. Our next piece of AI news. Uber Technologies is expanding beyond its ride sharing origins by entering the AI development outsourcing market with a new division called Scaled Solutions.
Jordan Wilson [00:19:15]:
Interesting name choice there. So the move leverages Uber's gig economy expertise to provide AI model training and data labeling services, potentially reshaping its business model and opening new growth avenues. So Uber's entry is timely as global demand for human vetted data to train AI models is increasing, highlighted by the success of companies like Scale AI. Right? So that's scale That's just kind of Scale AI's big thing, which is why I thought it was kind of, I'm guessing either strategic or ironic or maybe a little bit of both, that Uber decided to name this new division Scaled Solutions when the One of the leaders in this space is Scale AI. So, Uber is recruiting contractors with programming skills, language proficiency, and natural and cultural knowledge in countries such as India, the US, Canada, Poland, and Nicaragua among others. So contractors will perform tasks like image labeling and text annotation with earnings based on completed tasks, aligning with Uber's existing payment model. This is weird y'all. Let's let's let's be honest, but this is, I don't wanna get into this.
Jordan Wilson [00:20:33]:
I don't wanna get into this too deeply, but I'm on the record. My very first, my very first podcast like 2 years ago was AI is gonna take way more jobs than it creates, and I do think and I've said this before many times, traditional 9 to 5 jobs are gonna be gone in a couple of years. Right? Or at least look very, very differently or a much fewer percentage of people having 9 to 5 jobs. So I don't wanna get into the whole, like, AI is is gonna take all our jobs and and universal basic income. That's not what this is about, but I think a lot of us in a couple of years are gonna be having similar positions like this where you are working in your area of expertise to help big tech companies gain, you know, domain specific and niche knowledge. That's that's the future. I think we're all gonna be having, you know, maybe a handful of these kind of side gig freelance jobs in our area of expertise. Maybe in addition or maybe in lieu of, traditional 9 to 5.
Jordan Wilson [00:21:31]:
So I don't know. What do y'all think? It's it's it's kinda weird to think about. Right? And people thought I was crazy when I said this until Reid Hoffman, one of the co founders of LinkedIn said the same thing. And then everyone's like, oh, yeah. We should we should take take a look at this and, you know, relook this, you know, 9 to 5 culture and, you know, what full time employment even means and the future of it, etcetera. Right? When I said it, I was just weird. But when, you know, some of the smartest people in the world say it, they're like, oh, okay. Maybe we should think about this.
Jordan Wilson [00:21:58]:
But I mean this Uber news, it's not surprising. It's not surprising if I'm being honest. I think we're gonna see a lot of other big tech companies shift to something like this because the higher quality human data that we have to start with. Right? Because one thing that we've seen is unfortunately, we're seeing this cycle of AI regurgitation. Right? So so much of the quote unquote new, content going into large language models training sets is created from AI. So, you know, we're getting unfortunately this ugly cyclical, data regurgitation thing where, you know, that it's it's lower quality data going into or I should say lower quality new data that's going into the next version of model. So there is an increased emphasis on human created data. Right? So turning, all of our, kind of, knowledge into data but having humans do it, you know, with working alongside hand in hand, Lawrence large language models.
Jordan Wilson [00:23:01]:
So so it should should be interesting to see how this how this one shakes out. So, yeah. Fred is saying sounds like a security risk, maybe. One a Lee from YouTube says, to be fair you're not weird because you said the comments about jobs. That's good. I'm just weird for other reasons. Right? Who else talks about AI every single day? It's extremely weird. Alright.
Jordan Wilson [00:23:26]:
Our next piece of AI news, I'm excited about this one. So NVIDIA has unveiled Fugato, a groundbreaking generative model poised, I think, to transform audio synthesis by blending music, voices, and sounds in unprecedented ways. So this new model employs cutting edge synthetic training methods and inference level combination techniques, allowing it to create unique sounds that have literally never been heard before by human ears. Right? Such as saxophones barking or voices, you you know, singing classical music underwater. So, Fugato is not yet publicly available and NVIDIA did not really, release a lot of details about how, when or if they would make the model publicly available, but a sample filled website which we linked to last week in our newsletter and we'll link to again demonstrates its vast range of capabilities. So according to NVIDIA researchers, crafting a meaningful training dataset for Fugato posed challenges as it re required revealing complex relationships between audio and language. So the team utilized a large language model to generate Python scripts that create instructions for different audio personas, enhancing the model's ability to synthesize and modify sound traits. So existing audio datasets lacking inherent trait measurements were enhanced by using audio understanding models to create, quote, unquote, synthetic captions, providing natural languages, language descriptions that quantify traits like gender, emotion, and speech quality.
Jordan Wilson [00:25:04]:
So audio processing tools further further analyze clips, of an acoustic level, measuring aspects like fundamental frequency variance and reverb. Alright. So, I'm gonna try this here. We haven't done this in a while, but let's go ahead and take a quick listen. So live stream audience, if you could, let me know if you're hearing this. Sometimes technology doesn't want me, to share audio, but let's go ahead and take a listen here. Live stream audience, let me know if you can hear this, but we're gonna listen to about, 15 to 20 seconds, from this NVIDIA, kind of, output here. Let's go ahead and take a listen.
AI [00:25:46]:
Or direct immersive and shifting soundscapes for film or audio productions.
Jordan Wilson [00:25:52]:
Alright. So this is, the video in, NVIDIA released and here is the prompt that went or the input that went into Fugado, and then you can hear the output. So they don't read this. So it says create a sound where a train passes by and becomes a lush string orchestra. So you're gonna hear that and then, the next 2 that they describe, you should be able to hear them. Here we go. That's very impressive. Alright.
Jordan Wilson [00:26:34]:
Let's listen to 1 or 2 more here quick.
AI [00:26:36]:
Instructing Fugato to extract audio elements from a sound clip, such as isolating a voice track in a piece of music is just
Jordan Wilson [00:26:57]:
as easy. That is difficult to do. So, Marie Marie is saying, this is great. The train whistle turning into an orchestra. So let's listen to one more, but that was combining multiple prompts. So it it was combining or essentially multiple inputs. It was it was taking a flat music track, right, that had vocals and audio and it separated them, which anyone that has worked in audio or video editing knows that is very very hard and very time consuming. And up until generative AI has kind of been impossible.
Jordan Wilson [00:27:30]:
I think we're gonna listen to one more here.
AI [00:27:32]:
Fugado also allows you to generate new speech samples. Kids are talking by the door.
Jordan Wilson [00:27:37]:
So this input said, in a calm voice with an American accent, say kids are talking by the door. Alright. So let's listen to it, and then they're gonna say how they modify it.
AI [00:27:48]:
Kids are talking by the door. And if you want a different delivery, Fugado can do that too.
Jordan Wilson [00:27:55]:
Then it says, turn this calm voice into an angry voice. So, again, stat like using your output as your input and then modifying it with text. Ready?
AI [00:28:04]:
Kids are talking by the door.
Jordan Wilson [00:28:08]:
Super impressive. We haven't really seen anything like this yet. So Jessica asking, will this work to clean up background noise, on an old cassette tape? I think Jessica in theory, if NVIDIA releases it, that's what it can do with natural language. Right? And yes, I I do have quite a few contacts at NVIDIA. So, there's a couple, new NVIDIA, kind of concepts or models that I'm gonna try to get people on the show here. Suzanne on YouTube says, wow. Michael says, yes, there is the one that they changed the voice tone that we just heard. Alright.
Jordan Wilson [00:28:43]:
Let's get back to the AI news and we're gonna look at one more demo. So AI video models went crazy this week. So Luma Labs released their Dream Machine 1.5 update. So it enhanced realism and motion consistency. So this update also includes a mobile app. Users can also, in Luma Labs, now generate consistent characters. That's a big one. And it supports conversational interactions.
Jordan Wilson [00:29:10]:
So not just having to completely re prompt. So that's from Luma Labs Dream Machine 1.5. Then we got updates from LTX Video. So they released their tool as an open source option. Available on Hugging Face Spaces, allowing users to run their model locally, LTX video. Yeah. You will need a little bit of a power though. I think optimal performance they said if you want to run, this LTX video locally, which the fact that you can create AI text to video models locally on a machine is wild.
Jordan Wilson [00:29:47]:
Considering like a year ago, you you know, even what we have now which we would consider terrible a year ago, was revolutionary. Revolutionary a year ago. So you can actually create pretty decent, AI text to video, videos. But you do need at least, an NVIDIA RTX 4090 GPU, and I think that's like $2,000 for the chip. So yeah, you can run it locally but you gotta have a pretty powerful, chip, GPU there. And then last, last but not least, we have an update from Runway. So a couple things here. So one is their new expand video.
Jordan Wilson [00:30:26]:
So that's a feature that allows users to reframe existing videos into different formats. Such as as an example, going from a 9 to 16 or kind of a vertical video to a 16 by 9 and then filling it in and then doing it all over again. They also released an AI image generator called Frames. Right? So they're they're going about it the backwards way, kinda. Right? Starting with AI video, and now they released AI frames. So frames, it offers high quality photo realism and advanced stylistic control, enabling the creation of consistent aesthetics across different worlds. Alright. So, let's just quickly now, take a look here for our livestream audience at least.
Jordan Wilson [00:31:09]:
You can see this new, this new frames feature. So let's go ahead and take a look, and I will try to narrate it for our podcast audience. So it's starting here with a 16 by 9 video, right? And then with a prompt, you can turn that into a vertical video. So it's creating so it's this woman, you know, outside kind of looking, it's a close-up of a woman, presumably at a park or a waterfall but you don't know. So then it adds all this new vertical information and you see, oh yes, it is the top of a waterfall but that is all AI generated. So what started as a close-up kind of, kind of horizontal video turned into a vertical video. So now you have all this new AI generated information above this woman's head showing a waterfall. Then you can take that as an example and expand it even more.
Jordan Wilson [00:31:57]:
So you can just keep expanding as an example, going from vertical to horizontal and you add all these new things and then you can turn that large one into verticals. And then we'll see it turns it back, you know, back and forth. So the same thing here, you have this, the second example of a man taking photos, it looks like in a cave, a vertical video. Right? Turns it horizontal, so now you see the beginning of what looks like a little lake inside of a cave or something like that. And then you see this new information generated by AI, that's you know, kind of like the top, the top of the cave or light peeking in that was not previously there. So we go from, again we go from horizontal video to vertical at it with AI, and then again taking that and making it even more, you know. So going multiple times, it it really went from this close-up of this man taking this photo and you don't really know what it is. You're like, okay, it kind of assumes it's a cave and then it extends it, right, with this new generative extend and adds And you could just keep doing it more and more.
Jordan Wilson [00:33:00]:
It adds more detail, more visuals. This is a small feature that I'm personally pretty excited about. Yeah. Tara just said, wow. Shannon said, brilliant. Jackie said, great for various social social post variations. Great point, Jackie. Right? So if you shoot something vertical, right, and you wanna share that maybe on YouTube, you know, maybe you don't have something great.
Jordan Wilson [00:33:23]:
Well, you can use runway. Pretty pretty impressive, if you ask me. Alright. We got a lot going on, let's get to our last couple AI news stories. This one's weird. So a viral video, that just kind of silently dropped right around the Thanksgiving holiday here in the US, showed a small robot named Earby convincing larger AI powered robots to leave their posts, raising concerns about AI's influence on human like interactions. This is so weird. So the incident reported by Sky News occurred in August at a Chinese startups and was captured by closed circuit TV recording, but was only recently surfaced online.
Jordan Wilson [00:34:07]:
So this little AI powered robot, you know, working its job at this, Chinese AI startup approached its larger AI robot companions asking them if they were working overtime, to which the larger robot replied that it never gets off work. Then the small robot, Urbi, or Urbi, I don't know what its name is, then invited it to quote unquote, come home. Yes, there's actual transcripts. I don't know if this is a huge marketing stunt or what, or if this is just an eerie look into our future where AI robots are talking to each other. Anyways, Urbai convinced them and all of these other robots, I think there was 5 of them, literally, quote unquote left their job, right? Working here at this, factory and went home. So unexpectedly, other robots joined leaving their positions which researchers did not anticipate. Highlighting the potential oversight in AI programming. So the video has sparked discussions on social media with users humorous, humorously referencing movies like I, Robot and Terminator while expressing concerns about AI autonomy.
Jordan Wilson [00:35:14]:
Yeah. I don't even know what to say about this one. This is weird. It's very weird. It's very weird. But, yeah, if you wanna see that video, we'll be sharing it our new in our newsletter. Alright. So does NotebookLM have some competition at least when it comes to their AI generated podcast? Maybe so.
Jordan Wilson [00:35:34]:
So 11 Labs has launched a new feature called GenFM, which is available on the 11 Labs reader iOS app. So this enables users to create multi speaker podcasts by uploading different types of content such as YouTube videos, text or documents. So this innovative feature supports 32 languages including English, Hindi, Portuguese, Chinese, Spanish, French, German and Japanese, making it accessible to a global audience. So like NotebookLM's audio overviews feature where it creates a deep dive podcast, these 2 AI hosts, you can upload all of your content and it creates a pretty engaging kind of 2 speaker, kind of, piece of, podcast, right? This does the same thing. So GenFM from 11 Labs, it only works right now in their app but it automatically selects 2 voices from over a dozen available options so that kind of separates it or gives it a nice feature that NotebookLM does not have. Adding human like elements such as umms and ahs to create a more natural listening experience. So, yeah. I know these AI generated podcast summaries can add ahs and umms, but can they add random hot bit, you know, random rants like me? Can they drag on a new story that should take 30 seconds to read and turn it into 2 minutes like me? Right? I say that you know, kind of tongue and cheek but you know, maybe I should just be a little more concise here.
Jordan Wilson [00:37:09]:
Alright. Two last stories. We're saving some of the biggest ones for last. So, according to reports, OpenAI is considering advertising to produce, to boost its revenue. So OpenAI valued at a reported $150,000,000,000 is exploring advertising as a new revenue source according to the Financial Times. So the company aims to capitalize on its leading position in the AI sector with ChatGPT achieving over 250,000,000 weekly users. So OpenAI is aiming to reach 1,000,000,000 users by next year by building new data centers, reflecting its growth ambitions, following significant funding. So OpenAI's CFO, Sarah Friar, has emphasized a thoughtful approach to implementing ads, ensuring they align with the company's growth strategy.
Jordan Wilson [00:38:04]:
So despite rapid revenue growth to about $4,000,000,000 annually according to reports, OpenAI is facing higher costs, planning to spend over $5,000,000,000 a year in the near term. So, yeah, they may be losing money according to reports. But also, the company has been hiring advertising talent from tech giants like Meta and Google, indicating serious consideration of an ad model in the near term. So CEO Sam Altman is reportedly warming up to the idea of ads, though internal discussions continue on how they might be integrated. OpenAI's revenue largely right now comes from its API and ChatGPT licenses with the comp and the company is seeking more lucrative opportunities. So Friar noted that ad models can shift focus from users to advertisers but see unexplored potential in current business strategies. So the move towards advertising reflects broader trends in the AI industry where companies like Perplexity and Google are already piloting, AI ad models or ads in their AI, generate summations, AI search models, whatever you wanna call it. Alright.
Jordan Wilson [00:39:15]:
Last last but not least, Elon and OpenAI are fighting again. Why? So Elon Musk's attorneys have filed for a preliminary injunction against OpenAI, its cofounders, and Microsoft claiming anti competitive behavior. So, yes, this is brand new. If you're like, wait, I've heard you talk about this before. Well, this is just a preliminary injunction, a new one. So the motion was filed in the US District Court for the Northern District of California, and it accuses OpenAI and its associates of illicit activities and seeks to halt them. So allegations include discouraging investors from backing OpenAI's rivals like Elon Musk's own AI company, XAI. Right? Makers of the Grok chatbot that I don't know anyone that uses.
Jordan Wilson [00:40:09]:
So the filing also claims OpenAI's conversion to a for profit entity and self dealing transactions harm marketplace competition. So Musk's legal team argues that irreparable harm will occur if the injunction isn't granted aiming to preserve OpenAI's nonprofit character. The lawsuit revives Musk's previous legal battles with OpenAI accusing the company of straying from its original nonprofit mission. So Musk was obviously a co founder of OpenAI and left in 2018 over disagreements about the company's direction and then later founded XAI. And XAI, recently announced a $5,000,000,000 funding round positioning it as one of the most funded AI companies globally. So this inject, injunction in alleges OpenAI's actions deprive XAI's XAI's of capital by extracting commitments from investors not to fund competitors. So OpenAI is backed by Microsoft with a reported $13,000,000,000 investment, faces acquisitions of illegally sharing proprietary information and engaging in self dealing. So Musk's attorneys argue that maintaining OpenAI's nonprofit status is crucial to protect its founding mission and public interest.
Jordan Wilson [00:41:31]:
OpenAI spokespersons dismiss Musk's claim as baseless stating the lawsuit lacks merit. So I don't know. I'm not gonna go too far into this. I did an entire episode once that talked about this original lawsuit and just kinda I poked more holes in it than, you know, Swiss cheese at the grocery store. This doesn't make sense. What Elon Musk is saying makes absolutely no sense. And if I had to agree here, I'm obviously siding with OpenAI. Okay.
Jordan Wilson [00:42:01]:
Because Elon Musk is claiming that, you know, making these anti competitive claims and is essentially saying OpenAI is forming a monopoly. They're too big, too good, no one can compete, and we can't raise money. Right? Because what Elon Musk is saying well, you know, Sam Altman and OpenAI here are are asking investors not to invest in other AI companies, which we've reported on. So, I mean, they can. Right? At least from what we've seen in reportings, there's no legal agreement that says they cannot. At least the reporting we've seen has has said that OpenAI is asking its biggest investors not to invest in its rivals, which seems like common sense. Right? If you're cashing a $1,000,000,000 check from someone, you probably want to hope that that same, investor is not funding one of your rivals. But also, this doesn't make sense because Elon Musk is also, founding XAI in this Grok AI chatbot.
Jordan Wilson [00:43:07]:
Right? So on on the left hand, Elon Musk is saying Grok is is gonna be the most powerful AI model in the world. Right? And he's building all these, you know, compute gigafactories. Right? Or whatever he's saying with, like, a trillion NVIDIA GPU chips. Right? So on the left hand, he's saying, oh, Grok, like, r a I chatbot is is, you know, one of the best in the world and it's gotta be the most powerful. So the left hand, you're saying, like, kind of like, hey, we're the best. I'm the best. Grok is the best. But then on the right hand in court, you're saying, oh, it's anti competitive.
Jordan Wilson [00:43:39]:
No one can compete. So is Elon Musk's left hand talking to his right hand? Doesn't seem like it. And that's why a lot of people are writing off this lawsuit even though people are gonna be talking about it a lot. I don't know. I'd have to side with OpenAI and say this lacks merit. It's very legal for a non profit company to go through proceedings to convert to a for profit company. Or Yeah. Yeah.
Jordan Wilson [00:44:05]:
And the timing the timing of this lawsuit like Fred said, yeah. Particular, yeah, particularly interesting timing here. Michael just saying, OMG Musk, let it go, face palm. Shannon says, loves the rambling. Alright, well we will wrap the rambling up. So real quick here, the quick too long didn't listen version of AI news. So OpenAI's Sora AI tool was leaked after some artist backlash. Anthropic shipped a lot.
Jordan Wilson [00:44:36]:
They shipped their new model context protocol as well as their new customizable writing styles. We saw a new reasoning model from Alibaba's Quen team called qwq32b. Uber is apparently venturing into AI development outsourcing with its scaled solutions. NVIDIA released a new model called Fugado, a audio model that does a bunch of crazy things. The AI video companies went completely bonkers, over our kind of quote unquote Thanksgiving break here in the US with Luma Labs, LTX, Video and Runway all releasing a bunch of things. Speaking of releasing, a bunch of robots followed each other apparently out the door. So, you know, some weird things to worry about when it comes to AI robots talking and working with each other. 11 Labs released GenFM, offering a very similar feature or capability to NotebookLM's audio overviews.
Jordan Wilson [00:45:40]:
Open AI is considering advertising so we might be seeing, a bunch of ad results in ChatGPT soon according to reports. And then last but not least, Elon Musk is seeking a new injunction against OpenAI and Microsoft alleging anti competitive practices. Alright. That's it y'all. I hope this was helpful. If so, tell someone about it. Don't be a jerk. Share the love y'all.
Jordan Wilson [00:46:04]:
So if you are listening here on the live stream, go ahead and click that, repost button. Tag a friend, someone that needs to hear this. If you're listening on the podcast, please make sure you follow the show. You know, click that follow button or whatever that button says, and if you could leave us a rating, that would be great as well. Alright. Thank you for tuning in. If you haven't already, go to your everydayai.com. Like I said, we will be recapping not just today's podcast.
Jordan Wilson [00:46:28]:
We reference a lot of things, videos, you know, all these new releases. You can go read about them, see them, watch them. It'll all be in our newsletter, as well as on our website, your everydayai.com. It is a free generative AI university. You can sort all of our past episodes by category. Whatever you wanna learn, we have it there. Thank you for tuning in. Hope to see you back tomorrow and everyday for more everyday AI.
Jordan Wilson [00:46:50]:
Thanks y'all.