Everyday AI Podcast – An AI and ChatGPT Podcast - EP 384: AI News That Matters - October 21st, 2024

Episode Date: October 21, 2024

Are OpenAI and Microsoft fighting? Why does the U.S. military WANT AI deepfakes? Will Perplexity's new features make it a ChatGPT competitor? We bring you this week's AI news that matters. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on AIUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. Pentagon & AI Deepfakes2. NVIDIA's New AI Model3. Google Notebook LM Updates4. Adobe Max Conference and Related Updates5. Microsoft and OpenAI Equity Negotiations6. Apple's AI Development7. Perplexity's New Features8. Meta's AI DevelopmentsTimestamps:00:00 AI updates: Deepfakes, Adobe, Microsoft, OpenAI clash.04:36 Microsoft to launch customizable AI agents soon.09:10 Mac app for ChatGPT out before Windows.12:09 Apple's AI rebranding lags; partners with OpenAI.16:57 Perplexity: Good for insights, unreliable for citations.20:14 US government explores deep fake technology's implications.21:46 Pentagon wants AI deepfake users for understanding adversaries.26:10 Open-source model excels in specific benchmark test.30:07 Notebook LM launching business edition; future charges expected.34:20 Adobe video generator launches, beating OpenAI's release.37:10 Microsoft, OpenAI seek bank help for negotiations.38:47 OpenAI restructuring for profit and societal balance.44:15 Microsoft offers ChatGPT app for paid users.Keywords:Pentagon, AI Deepfakes, The Intercept, NVIDIA, AI Model, Llama 3.1 Nemo Tron 70b Instruct, Google, Notebook LM, Gemini 1.5 model, business edition, AI audio, Adobe Max Conference, Adobe Firefly video model, Adobe Firefly 3 Image Generator, generative extend feature, Microsoft, OpenAI, equity negotiations, AGI, Microsoft-OpenAI partnership, Copilot, GPT technology, Satya Nadella, Sam Altman, Jordan Wilson, webinar, podpp.com, Apple AI development, Siri, NLP capabilities, Apple Intelligence, Perplexity, US Military, deepfake technology, ChatGPT, Customizable AI agents, Dynamics 365, Meta, synthetic data, Agentforce.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. The U.S. military might want to start using deepfakes.

Adobe just dropped a ton of new AI updates. Speaking of drops, minutes ago, Microsoft just announced when it plans to drop its autonomous AI agents. And then speaking of Microsoft, are Microsoft in Open AI fighting? y'all it's been a wild week in the world of AI news literally every single big tech company is out there making pretty big headlines and we're going to cover them all today in about i don't know 20 or 30 short minutes all right what's going on y all my name's jordan wilson and welcome to everyday AI this is your daily live stream podcast and free daily newsletter

helping everyday people not just keep up with what's going on in the world of but how we can all use that information and get ahead to grow our companies and to grow our career. So that might sound like you, right? Like, oh, I want to keep up with AI. I want to get ahead. I want to be the smartest person in AI and my company. Well, you are in the right place.

So if you have not already, please go to your everyday AI.com. Sign up for the free daily newsletter. You know, on Mondays, we usually do this show, the AI news that matters, keep catching you up on everything in the world of AI. Maybe you just listen Mondays, but we always have. a great lineup of other experts coming in from all aspects of business to teach you about AI. So make sure you tune in every day if you can. All right. So I'm excited to get today's show going. Real quick, I just got to shout out everyone joining us live. I'm in a new,

new little situation here, not recording in my normal Chicago studio, so to speak. So thank you for Michael, Rolando, Brian, Kurtz, Tara, Joe, everyone else, Chris, for joining us live. So without further ado, let's just jump straight into the AI news that matters because like I said, there is a ton this week. All right. Let's start with meta, shall we? So meta has just released a new AI model, including a self-taught evaluator to reduce human involvement. Yeah, yay. Less humans, more AI.

So META has just announced the release of several new AI models from its research division, highlighting a significant shift toward less human intervention in the AI development process. So the newly introduced self-taught evaluator model is designed to assess AI responses with minimal human inputs, potentially leading to more autonomous AI agents in the future. So this model uses a chain of thought technique. similar to that use in Open AI's recent 01 models, and it breaks down complex problems into manageable steps, enhancing accuracy and challenging subjects like science, coding, and math.

So meta's researchers train the evaluator model entirely on AI generated data. Yeah, so essentially, let me put this in human talk. So this new AI model from meta is made to fact-check other AI models, and it was made with synthetic data, or data created from AI. Yeah, if that's not meta from meta, I don't know what is. So researchers involved in the project expressed hope that as AI systems become increasingly more advanced that they will surpass human capabilities in self-evaluation and correction.

Other companies such as Google and Anthropic are also exploring similar concepts, but typically do not release their models for public use, setting meta apart. in this regard. So yes, this is an open source model that we just saw from meta. So, y'all, this is pretty strange, right? Very cool, right? If you're an AI dork like myself, maybe you are. But the fact that this new model self-evaluates other AI models and it's trained on synthetic data made by AI, right? It's hard to almost wrap my head around this early in the morning. All right. Wrap your head around this. This is hot off the presses. So Microsoft has just announced a launch date for customizable AI agents. So this is going to be coming in the public preview next month.

So Microsoft has just announced that it will allow businesses to create their own autonomous AI agents starting next month, ramping up competition against Salesforce recently introduced AI tools called Dreamforce, or sorry, that was announced at Dreamforce. It was called Agent Force. So, yeah, Salesforce had this pretty big pivot last month at their Dream Force conference when they said, hey, we're kind of an AI agents company now. And they introduced Agent Force. So Microsoft just now revealed some more of its plans to transition its autonomous agents

in co-pilot studio from the private preview. it is currently in, only available to a very select few number of people, to a public preview next month, so starting in October, allowing more organizations to build their own AI agents. So these AI agents, well, they can do kind of anything, right? They're designed to function as virtual workers capable of performing various tasks independently, representing a shift from traditional chat interfaces to this more integrated solution. So just a couple minutes ago, actually, at the recent AI tour event in London, Microsoft

showcased an example of these new autonomous AI agents with the consulting firm McKinsey. So illustrating how an AI agent can analyze emails, retrieve relevant information, and summarize responses, all just using natural language. Yeah, you don't got to be able to, you know, spit a bunch of structured inputs, or code to these agents. You just talk to it like, hey, agent, go do this work. So Jared Spataro, Microsoft's corporate vice president, highlighted the potential business value of these agents,

noting that in this example, McKinsey reported a reduction in lead time by as much as 90% with their AI agent. Yeah, everyone's always talking like, oh, is there a return on investment? Like, what's the actual time saved? Well, I don't know. maybe McKinsey and Microsoft are two names that you've heard of before. 90% time savings.

Let that one sink in as I take a sip of water. So Microsoft plans to induce 10 new autonomous agents in its Dynamics 365 suite, which includes applications for sales, services, finance, and supply chain management. So like I talked about, this isn't. just Microsoft and Salesforce, right? This is kind of everyone. We've heard reports that OpenAI, Google, DeepMine, just about everyone in the AI space is starting to heavily invest in AI agents.

So Microsoft actually here is probably going to beat Salesforce to the first major company releasing agents to the public. So, yeah, wild times. y'all. I'm not sure if you all have seen the examples, but we'll make sure to share this in our newsletter today. So yeah, if you haven't already, make sure to go check that out. Another thing to check out, well, if you are a Windows user, OpenAI has just launched ChatGPT. It's ChatGPT Windows app for select users. So OpenAI has just begun previewing a dedicated Windows app for its chat GPT platform signaling a significant shift in making AI more accessible to its users.

So currently, this early version of the app is only available to paid users. So if you have a chat GPT plus, Teams, Enterprise, or EDU accounts with plans for a more comprehensive experience later this year. So you might be wondering, okay, well, why do I need a chat GPT app inside? of Windows. Well, there's a lot of advantages and benefits. One is it's just a better experience, right, than having a thousand tabs open. If you really want to get the most out of chat, GPT, I use personally, I love the Mac app. I'm on the Mac app right now. So, you know, originally we heard back when they announced this that it was only going to be coming to Mac.

So that's also an interesting, an interesting thing to note here that they did end up coming out with this Windows desktop version. So yeah, if you haven't already checked it out. So it's been out the desktop version for Mac has been out since June. So Windows users kind of got the short end of the stick here, having to wait an additional four or five months for the Windows version. But in our newsletter today, I'm sharing a video that I already created that kind of shows you the benefits of using the app versus using the

browser. All right, so make sure to check that out if you are interested in the chat GPT app for Windows. Hey, live stream audience, do you guys prefer curious here? Also, I'm on the road. So normally, I've usually had one or two coffees by now. And it just hasn't hit me yet. The water's just not hitting as hard as the coffee does. But hey, live stream audience, do you listen or sorry, Do you watch or use chat GBT in the desktop app or do you use it in the browser? I'm always curious how other people use chat GBT. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience.

Meet Firefly AI Assistant now live in the Adobe Firefly app, the all-in-one creative AI studio. Powered by Adobe's creative agent, Firefly AI Assistant lets you start with, your vision, just describe what you want, and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks like batch editing photos, creating mood boards,

portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com. Apple is far behind. So recent insights from Bloomberg reveal that some Apple employees believe the company is approximately two years behind in artificial intelligence development, highlighting the competitive pressure in the tech industry. So at WWDC 24, which happened back in June, that's when Apple first officially announced Apple intelligence.

So this is Apple's new, their attempt to rebrand AI, right? Don't know what that's all about. But that was kind of its first official foray into the generative AI and large language model landscape, at least bringing it to all of Apple's devices. But a recent report from Bloomberg looked at internal employees saying that, yeah, we're up to two years behind. And I've been saying this since the get-go. So there was a lot of reports last summer that said Apple was spending millions, with an S, millions of dollars a day on training models internally.

Yet when they launched, really one of the highlights was of Apple Intelligence was they decided to partner with Open AI and use its chat GPT model for more advanced kind of questions from users. So if you do ask a simple question, there's a good chance that the Apple intelligence will use its own Apple-made kind of edge or small language model that lives on your device. However, it seems like the majority of more complex queries are actually just going to get passed off to chat GPT. So, yeah, Apple has been facing a ton of scrutiny over the last couple of months, especially as Apple intelligence, just started rolling out in beta to a lot of users. I did a couple reviews both here on the show, and we reviewed a couple more features in-depth on our YouTube channel. I was not impressed, right?

So not only was that, you know, I think that whole rebranding of Apple, artificial intelligence and calling it Apple intelligence, kind of cheesy, if I'm being honest, but the features were not that good, right? These are all features that we've had for one year, two years already. So it's really to say that Apple is two years behind, you know, a lot of times you see these reports. And, you know, if it's unnamed sources or anonymous sources, you're like, I don't know if that's true.

This one, I 100% believe this report from Bloomberg talking about that even Apple's own employees are saying they are multiple years behind the other big tech companies. But the Bloomberg article did mention that Apple does have a little. bit of a history of starting slow and catching up as an example. Apple Maps when it first came out was very bad, right? And now it's pretty good, right? There's actually features inside of Apple Maps that I would say are better than Google Maps. And that's just one example that Bloomberg noted in its report in terms of Apple, maybe starting very slow, but eventually catching up. Also, Apple right now plans by 2026. Yeah, not 2025, but by 2026 to have fully rolled out its Apple intelligence, including the revamped Siri that is supposed to have that more natural language processing

similar to chat GBT's advanced voice mode. So yeah, by 2026. Oh, I do not want to be working at Apple right now. Yeah, they are so far. behind. All right. Perplexity, more AI news. Y'all, I told you. This week was a who's who. Every single name has been making a splash this week.

So perplexity has unveiled some new features to enhance research capabilities for users. So perplexity has just announced the launch of two new features, internal knowledge search, and it's updated to spaces, which aims to provide users. which aims to provide users with greater flexibility in sourcing information. So the new internal knowledge search feature allows users to search both the web and their team's files, enabling multi-step reasoning and code execution for more comprehensive responses. Previously, users could only upload personal files, but this new tool expands the AIS capabilities to include internet searches, providing a more integrated,

approach to search. The Spaces feature serves as an AI-powered research and collaboration hub, allowing users to customize Perplexity's AI assistant to meet specific project needs and data requirements. So let me simplify what's happening here. So largely, I think people have viewed perplexity as an Answers engine, right? An AI-powered Answers Engine, a competitor to traditional Google search, which I think it is. I will say this. Perplexity in my experience has a lot more hallucinations than people talk about. Right. So as great as perplexity is, I think it's only right now very useful for certain business use cases. I still think, you know, when we always talk about human in the loop, I think you actually have to be very careful, especially if you are relying on

perplexity to cite specific facts or sources. I think perplexity does an amazing job at being able to grab general ideas and insights from many articles online at once. So I think it's great for replacing Google search. But I don't think right now it's great at producing reliable citations, sources, and facts like you would hope for out of a large language model. So something that people overlook about perplexity is, well, it actually taps into, other large language models. Although perplexity does have its kind of in-house sonar, which is just a fine-tuned version of meta's models,

for the most part, people use Claude 3.5 Sonnet or OpenAI's GPT40 as the actual engine that powers it. So what I think is happening here with perplexity, they are trying to compete more directly with those companies, right? With OpenAI, with Anthemps. Clawed with Google Gemini with these new features, these new internal knowledge search and the spaces feature. But similarly to how you can upload your files in GPTs inside of chat GBT and inside of gems

in Google, I won't even put projects from Claude in this category because with Claude projects, you can't query the internet. So I think what they're trying to do here is they're trying to give business users specifically an option to bring in their own knowledge base, essentially, upload files, but then at the same time, get the good stuff out of perplexity, which is the ability to really browse the web very quickly and to bring you insights. So yeah, let me know if you guys want to see, you know, maybe should we do an updated deep dive on perplexity with all these new features?

Let me know in the comments if you want to see that. Speaking of things you want to see, this one's weird. All right. So according to reports, the U.S. military is seeking advanced deep fake technology. Yeah, they're looking for this. They want to use it. So the United States Special Operations Command is reportedly actively pursuing the development of deep fake technology to create convincing online personas that are indistinguishable from real humans. So the Department of Defense's Joint Special Operations Command has released a procurement document

outlining a desire for technologies that can generate realistic online profiles for social media and other platforms. So the Intercept actually had a great article breaking all of this down that we'll link to in our newsletter. But according to the report, the goal is to create user profiles that appear human, complete with multiple expressions and high-quality identification photos that do not exist in reality. So the document that was released specifies the need for not just static images, but also video and audio elements, including selfie videos set against believable backgrounds to evade detection by social media algorithms. The push for deep fake technology reflects a broader tension, within the U.S. government where some agencies aim to maintain public trust,

while others are seemingly wanting to invest in better understanding and potentially using some of this deep fake technology. The implications of this are obviously huge as the implications of these technologies extend beyond military operations, raising questions about the future of information, integrity, and trust in digital communications for the general public. So yeah, this is one we're going to be keeping a close eye on. And here's the reality. I read a lot of AI news, right?

Every single day, we start the podcast with two or three of the biggest AI news stories. And then we dive into the topic for the day. And very rarely do I read a story and say, wait, what? This is one of those, this report from the Intercept talking about how the Pentagon actively wants to use AI to create deep fake users online. So yeah, there's not a lot of information out yet about the intent, as in why, right? You hope that they want to use this technology so they can better understand how our kind of adversaries across the world are using it, right?

So I get that piece of it, right? because other countries are trying to use this technology to maybe hurt or harm the United States. So from that aspect, I understand it. You really have to be able to use and understand the technology to prevent others from using it against you. But man, I wish there would have been that kind of reassurance, right? That is my assumption. So I wish in this report and in this paper that was kind of released, I wish that that would have come to light a little. little more explicitly, a little concerning.

Right? I don't know. Does anyone else, anyone else see that story? Just like, ugh, just me? Not a big fan. All right. Something I am a big fan of new AI models all the time. So, Invidia just launched a new AI model that outperforms some of the bigger companies.

So, Invidia, yes, the chip maker. has released a new model based on Lama's 3.1 from meta. So, Nvidia has unveiled a new AI model that surpasses offerings on some benchmarks from industry leaders like OpenAI Anthropic and Google, marking a significant develop in the AI sector. So this release is pretty noteworthy because obviously Nvidia is the leader in GPUs, which all of these companies need to train their models. Yet, InVidia just fine-tune meta's Lama model,

and it is already outperforming some models from OpenAI, Claude, and Gemini on certain benchmarks. So the new model is named Lama 3.1 Nemotron 70B Instruct. Great name. That just rolls off the tongue really early on a Monday morning. All right. And it's been introduced on hugging face. If you just want to go use it, we'll have that link in our newsletter today as well.

So like I said, the fine-tuned model is based off of Lama's 3.170B model, which is actually wild. Yes, these models take time, but meta's Lama 3.17B is not only the middle model of the 3.1 series. There is also the 405B. So that is, you know, 70B is trained on 70 billion parameters. 405B trained on 405 billion parameters. So this new model, which got some benchmarks higher than OpenAIs GPT40 and Anthropics 3.5 sonnet was on number one, the medium-sized model from Meta, but their old model, right, 3.1.

So a couple of weeks ago, meta just released their new model. Lama 3.2. So very interesting. So let's talk about the benchmarks, right? Because everyone's talking about this. And I think there's a lot of incomplete information. I think people were kind of losing their noodles. This went viral online last month for a couple of reasons.

Number one, it's Nvidia just fine-tuning a open source or open weights model from meta. But, you know, invidia's in hardware. So why are they getting into software in AI models? So that's number one. but it was just one specific benchmark under one specific kind of set of criteria. So I wouldn't pay too much attention to this. Obviously, I know a lot of people are talking like, oh, you know, Open AI and Anthropic, their new models, they're not even good because look at this new Nvidia fine tune.

It's way better. No, it's not. Just one. So the one very specific benchmark was the Arena Hard Auto on GitHub. and that was with no control style. That's like some super technical dorky stuff. But essentially think if you do 100 tests and all these tests have different conditions, it just outperformed on one with an 85.0 score,

which was higher than Open AIs GBT40 and Anthropics 3.5 sonnet. So here's the thing, y'all. I wouldn't pay too much attention to this, Although it is very cool, and yes, you can go use it. And it is very impressive for a 70 billion parameter open source model that it can perform at a pretty high level, right, of these close source proprietary state-of-the-art models. But for general use cases, that's not where it's at. Even Nvidia emphasized the importance or invidia caution that the model may not yet be suitable for specialized domains, such as legal, mathematical reason. etc. So, you know, even at Vida said, hey, you know, this essentially was fine-tuned and we were

able to get great benchmarks in certain aspects, but probably not a model that you want to start using every single day. A model I'm using every single day, literally, Notebook LM. So we covered this last week, middle of the week, but Notebook LM has announced some new features. So they've announced a new business addition in also customizable AI audio. features. So Google's notebook L.M, which is based on its Gemini 1.5 model, announced some pretty significant changes to updating or customizing the AI audio overview. So that's, you know, these AI podcasts that have gone viral over the last couple of months. That's from Notebook LAM. But before you kind of just, you know, would upload all of your source documents.

and click, generate an AI podcast. And then the deep dive, it's essentially two AI hosts. And they would come and have a very real and very captivating conversation based on all the source material that you upload. But it was just kind of a crapshoot, right? Because you couldn't really control what the host talked about. It was really up to their own, I'd say, agency, right? That's kind of weird to say.

But now you can customize it. So Google kind of unofficially called this, you know, pass a note to the audio presenters. But essentially it gives you about 100, about 150 words of custom instructions that you can upload beforehand. So we did a very fun show on this last week. We did it live. So the example that we did is we uploaded a bunch of documents, right? So if you haven't heard of notebook LM, it's fantastic. I personally love it.

Aside from chat, GBT, it is the AI tool that I use most. but essentially you can upload all kinds of different source documents in different ways. So we uploaded some YouTube videos, websites, I think PDFs on how to play cribbage. And then we uploaded some information on project management, right? So previously, if you created an AI overview of that, you wouldn't get anything that worked great. Essentially, they would say, okay, well, I think that this maybe project management piece was a mistake. So in our example, we put in some customized instructions and we said, hey, we're trying to teach project managers how to play cribbage, right? So we were able to kind of, you know, quote unquote, pass a note.

And the audio was actually fantastic. We shared the downloadable audio in our newsletter. And I think it actually turned out very, very well. So that was one new feature. The other new update from Notebook LM is they are launching a business edition pilot program. So I signed up, haven't gotten access to it yet. But we do assume in the future Google will be charging for Notebook LM. Right now it is free, which is wild because it operates as a grounded model.

So it only works with your data. So it really reduces the amount of hallucinations and increases the amount of trust. So it is a grounded model. It is retrieval, augmented generation, right, in a very simple form. and it's available for free. So Google kind of, you know, I think it's great that Google in Notebook LM is coming out with this business edition feature.

But what that means is probably notebook LM is not going to be free forever, which is probably for the best because they have millions of users. And I assume it is extremely expensive. So I think more than anything, this is par for the course, but bringing some more advanced features to teams. So once we get more updates on, the business edition will probably be letting you know. All right.

Two more stories, y'all, I told you this week, we got a ton. So Adobe Max. So Adobe unveiled a ton of new AI tools at their Adobe Max conference. So their recent announcements at the Max conference had the potential, I think, to really transform the use of AI in creative spaces, especially with some video and photo editing capabilities. So I'd say probably the biggest announcement out of all of this was Adobe officially introduced the Firefly video model. So yeah, like most things in AI, it's not publicly available yet. So it's a sign up page, right? Blog post and the wait list. However, the Firefly video model is a new

generative AI tool that allows users to create videos from text or images now available in, like like I said, limited beta rolling out very slowly within both the Premiere Pro app from Adobe and their Creative Cloud, as well as the Firefly web app. So yes, Adobe does have a dedicated Firefly web app for its AI generated photos and now its AI generated videos. So you might be thinking like, okay, well, why? And is this kind of late? Well, sure, right?

We have a lot of text to video or photo to video products. you know, as an example from runway, from a Luma Dream Machine, from Kling, right? There's a handful of just high quality AI, you know, text to video, photo to video products. But one thing that Adobe is is really, really pushing here is that it uses data that it says it has access to. Where a lot of these other AI video models, they don't really say how their model was trained, right? all these other big companies, they're kind of coy, right, on what goes into their models, right? The assumption is most of these models just have scraped the internet and have scraped sites like YouTube, which is using copyrighted works, right?

So Adobe, at least saying that it does not scrape the internet for its training data. And it only uses, you know, to create the Firefly models, it only uses, visuals that it has essentially rights to use. So pretty interesting there. Also on the photo side, Adobe did announce its new Firefly 3 image generator. Pretty good. I'd say it's better than Dolly, but still very far behind industry leaders such as Mid Journey, probably stable diffusion as well. And maybe even the Imagine model from Google that they just rolled out two weeks ago inside Google Gemini. So I do think instantly Adobe's AI photo generator, it's at least, you know, on par,

maybe not Tier 1A, but maybe Tier 1B. But the big thing here is when are people going to get access to the video generator? The quality looks really good. And they beat Open AI SORA to the punch, right? That's another huge thing. It's been about eight months since Open AI previewed its text video model SORA, but still don't have it, right? So at least Adobe here beating them to the punch.

The other big kind of announcement that I think, even though it's only two seconds, I think the new generative extend feature inside of Adobe's video editing programs is huge. So this is a new feature that essentially, and we showed this live last week as well on the show, but you know, think if you're a video editor and maybe you don't know, but a lot of times there's gaps. There's these small gaps in video. And what that means sometimes is, you know, editors will sometimes just reuse a lot of shots or they'll use a lot of slow motion

or some B roll that's not very good, right? So this new generative extend, you need to have at least three seconds of video, right? So anywhere from three to however many seconds, but minimum three seconds. And then you can extend it and you can add two seconds on the back end, which doesn't seem like a lot. I get that. And you're like, all right, Jordan, why are you talking about two seconds of AI video, well, it works with your video where so many of these other AI video generators

work with text and they work with photo. So Adobe bringing this generative extend into professional workflows, right? So many of the TV shows that we all watch, so many of the movies that we all watch, documentaries, et cetera, they're edited inside of Adobe. So I do honestly think this is going to. to be the first time that the majority of people in the world, right? Yeah, we all like, if you're listening to the show, you live in an AI bubble with me, right? We've all seen all these other AI video generators, but no one else has.

I do think this little two-second generative extent is going to be the feature that really brings AI video to the masses. All right. Our last piece of AI news, and we save the juiciest for last. So according to reports, Microsoft and Open AI are navigating some complex equity negotiations amidst Open AI's soaring valuation. So Microsoft's substantial investments in OpenAI are now totaling more than $14 billion since 2019.

And they're now at the center of some intricate negotiations regarding equity as OpenAI has transitioned from a nonprofit to a for-profit corporation. So according to reports from the Wall Street Journal, Microsoft and Open AI have engaged some outside help to help them talk through some of these tricky negotiations. So they've engaged Goldman Sachs and Morgan Stanley, respectively to assist them in determining Microsoft's ownership stake in the new for-profit structure and potentially renegotiate.

negotiating terms of their original agreement. So at least reportedly, the original agreement was open, was Microsoft invested at that time more than third or it was reported between $10 billion to $13 billion for a reported. I keep saying reported because it was never made public, but a 49% equity stake in Open AI. But now Open AI's valuation has skyrocketed to $157 billion. following a recent funding round with notable investors like Nvidia, SoftBank, and just about everyone. So Microsoft's earlier investment of $10 billion in January of 2023 valued OpenAI at $86 billion. So yeah, huge. That's about an almost doubling valuation in about 18 months. So despite the high valuation, Open AI projects a 5 billion.

million loss this year with revenue expected to reach 3.7 billion, but its expenditures and what it's spending are outpacing that. So that's why we have to look at this restructuring and what's going on here. So the restructuring of Open AI right now from a not-for-profit to a for-profit to public company aims to attract more investors while maintaining a commit to its original societal goals. So Open AI will still have a nonprofit that will continue to exist and retain a minority stake in the for-profit entity. So yes, the nonprofit charity is going to have an equity stake in the new for-profit, the new

for-profit kind of parent company. So CEO Sam Altman is expected to receive a larger equity stake in the new company, addressing previous concerns about his limited investment. exposure. So according to all of these reports, there's another significant concern is a clause in the contract that may restrict Microsoft's access to Open AI's technology if Open AI later develops essentially artificial general intelligence or AGI, right? So HEI is essentially when AI is smarter than every single human in every single task, right? So this clause, and we actually did an entire, like, hour-long show investigating this about three or four months ago.

So this one clause is designed to prevent potential misuse of the technology by Microsoft, according to reports, raising questions about how AGI is actually defined and who actually determines its arrival, right? Because, yes, the terms of this agreement change drastically. Once essentially, AGI has quote unquote, been achieved. But right now, it is Open AIs Board that has the authority to decide when AGI is achieved, but its CEO, Sam Altman, acknowledges that this assessment will be subjective and potentially ambiguous. So Altman previously noted that as progress continues, determining the

timeline for AGI becomes increasingly complex, suggesting it might be a gradual transition rather than a clear milestone. So yeah, this is going to continue, right? And I think we've really overlooked how monumental this Microsoft and Open AI partnership has been for both companies, right? So from a financial standpoint, makes a lot of sense for Open AI, right? One of their biggest investors also when we had the whole, you know, Sam Allman firing fiasco back in November and, you know, outsiders were saying like, oh, is open AI on the cusp of

collapse? I was like, no, but it was actually Microsoft CEO, Sadia Nadella, that reportedly helped lead talks that, you know, brought Sam Alman back into the fold and reinstated him as CEO. So yeah, there's a lot of these things under the hood that a lot of people aren't looking at, you know, another one is so much of what Microsoft is leaning into as in Enterprise is its co-pilot, right? Almost everything that Microsoft talks about now is based on co-pilot, which I think rightfully so. That's the AI is the future of work, right?

But co-pilot and essentially Microsoft's future systems, at least right now, are based on open AI's GPT technology, right? GPT4 or GPT40. So obviously Microsoft has made some substantial investments over the last couple of quarters to really bolster its own in-house large language model and small language model development to have less of a reliance on open AI and its GPT technology. So yes, these reports that just came out from the Wall Street Journal, which we will be linking to in the newsletter, pretty significant. Yes, this is standard.

Right. Anytime, you know, there's billions and billions of dollars invested and, you know, corporations, you know, switch their operating structure and, you know, so this is normal, but it is definitely worth paying attention to, as we'll definitely be paying attention to, on this show, as this is kind of this power partnership, not saying anything shaky, right, not saying they're on shaky grounds, but I mean, anytime you're bringing to the table outside parties to help you renegotiate a multi-billion dollar investment that really powers how we all work, it's definitely worth paying attention to. All right, y'all, I know that was a lot.

Here is the very, very fast recap of all of the AI news that matters this week. So meta has released a new AI model that literally just fact-checked other AIs, and it was built with AI powered synthetic data. So Mata from META Microsoft launched its new customer, or is launching, its new customizable AI agents for businesses in public preview next month. So that they just announced at their recent AI tour events in London. So big news there for Microsoft. Microsoft users now also getting the chat GPT desktop app if you are on a paid plan.

Our next piece of AI news was looking at a Bloomberg report about how Apple is reportedly, according to its own internal employees, at least two years behind, two years, behind in AI advancements compared to its competitors. Perplexity, our next AI news. Perplexity unveiled some new features. It's internal knowledge, search, and spaces, which I think are putting them more on path to compete with players like OpenA.I and Google Gemini. Next, the U.S. military is seeking to actually use some advanced deep-face technology.

that one's weird, but definitely worth paying attention to. Nvidia launched a new AI model that outperformed open AIs and Claude's models, but only on very specific benchmarks. And this was based off a Lama 3.17B. So, InVidia, they're fine-tuning a model. Notebook LM announced a bunch of new features, customizable AI audio, as well as a new business edition that they just launched Polyette 4, Adobe, Max, AI, everything at their AI, or sorry, their Max conference.

So including the Firefly video as well as the Firefly Photo 2, generative extent, I think is going to be a big one. And then last but not least, some new reporting coming out, kind of detailing this now maybe tumultuous relationship between Microsoft and OpenAI and some things that are going on behind the scenes there. that's a lot. I need a drink of water. I hope this recap was helpful. Every Monday we go over the AI news that matters. If this is helpful, please join us every other day where we bring in expert guests from all realms of work to help us all understand AI better. So if you haven't already, please go to your everyday AI.com.

If this was helpful, don't keep this as your little secret. Share this with your coworkers, families, friends, your dogs, neighbors, babysitters, dog walker, whatever. that means. Thank you for joining us, and we hope to see you back tomorrow and every day for more everyday AI. Thanks y'all. Meet Firefly AI Assistant, now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stay in control with the ability to step in and refine at any time. See it today at

firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.