Everyday AI Podcast – An AI and ChatGPT Podcast - EP 418: AI News That Matters - December 9th, 2024

Episode Date: December 9, 2024

No joke.... this has been the busiest week in GenAI news. Ever.  Amazon -- releases frontier models. Meta -- brings us a new Llama. OpenAI -- new models and features Google -- shipping AI literally everywhere  What happened? Why is all of this happening now? We'll dive in, and make you the smartest person in AI at your company. 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. Amazon AI Developments2. Eleven Labs Voice Agents3. Microsoft AI Developments4. Google AI Model Updates5. xAI Updates6. OpenAI's Latest Releases and Plans7. Meta's Llama 3.3 Model6. OpenAI-Microsoft RelationshipTimestamps:03:00 OpenAI o1 Pro: Elevated AI, exclusive, costly.08:14 Copilot Vision provides insights, prioritizes privacy, feedback-driven.11:58 Gemini surpasses OpenAI GPT-4 in leaderboard.13:58 Google DeepMind outperforms ENS weather, AI advancements19:47 Amazon's model surpasses OpenAI's context window.22:27 Amazon quickly reaches top-tier model status.24:33 11 Labs platform: multilingual AI for customer interaction.29:12 Musk criticizes OpenAI, joins government, impacts technology.33:24 OpenAI discusses removing AGI access clause with Microsoft.34:51 OpenAI's redefined AGI criticized by Elon Musk.40:47 A sneaky release of a semi-open model.44:51 Advanced voice mode updates, some features rumored.46:47 OpenAI announces operator preview, waitlist expected.Keywords:Jordan Wilson, everydayai.com, Amazon Frontier model, ChatGPT, AWS, Nova Canvas, Nova Reel, Anthropic, Eleven Labs, David Sachs, Microsoft Gemini Live, Microsoft Copilot Vision, Google Gemini 1206, Google Gencast, Google Veo, Google Genie 2, Sundar Pichai, XAI, Elon Musk, OpenAI o1 pro model, ChatGPT Pro, AGI definition, Meta Llama 3.3 model, OpenAI-Microsoft relationship, OpenAI public benefit corporation, OpenAI restructuring, AI regulations, Department of Government Efficiency, Large language models, AI developmentSend 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. What happened?

It seems like every single big tech company in the world decided to release new large language models over the last couple of days. I mean, we saw new releases from OpenAI, from Google, from meta, and Amazon finally through their kind of hat in the ring as well. well. So, you know, there's this great masterful film called Anchorman. So to quote it, like that escalated quickly. All right. What's going on, y'all? My name is Jordan Wilson. And welcome to Everyday AI. This is for you. This is your daily live stream podcast and free daily news that are helping us all, not just keep up with AI, but how we can use it to get ahead and to grow our companies and to grow our careers. If that sounds like you, bet you are in the right place.

Yeah, we do this Monday through Friday every single day. But on Mondays, we go over the AI news that matters. So you can walk into your office as the smartest person in AI. That's what I spend my weekends doing. So you can walk in prepared to grow your company, your career, your department, whatever it is. And another great way to do that is to go to your everyday AI.com. So right there, you can see.

sign up for our free daily newsletter because yeah, we have the live stream of the podcast every single weekday. But the newsletter is where you can actually take all the insights. So yes, we give you the normal what's happening in the world of AI. But when we have great guests on the show and, you know, when we dive in deep into all these topics, we actually recap and write new content for the newsletter on each day's topic. So you have to make sure to go check that out, as well as you can go listen to 420 plus now episodes of everyday AI. You can go read them, watch them, listen to them on the website. You don't have to go anywhere else.

It is your number one stop for AI information. All right. Also, join us tomorrow. You know, sometimes we do a little hot take Tuesday here on the show. So tomorrow, you are going to put me on the hot seat. Your questions. Already posted the show if you're on LinkedIn. it's in today's show notes.

So if you're a normal podcast listener, make sure to go click that. Get your question in now. You don't have to join us live. I know not everyone can, but get your question in now. What is your biggest question on AI? Grill me.

Put me on the hot seat. We're going to do that tomorrow. All right, enough chit chat, y'all. Let's get into the AI news that matters for the week of December 9th. Open AI. Yeah, 12 days of shipmiss, 12 days of open AI, whatever you want to call it. But we already got two big releases and two big updates. But the biggest ones, well, new models and a pricing strategy.

So Open AI has introduced the 01 Pro model and chat GPT Pro, marking a significant update to its AI offering. So the chat GPT Pro plan is priced at $200 per month, raising questions about accessibility and value. So the 01 model is different from the normal GPT4. So this is a reasoning model. It's a model that kind of thinks, does some kind of chain of thought thinking under the hood. So it is now described as the smartest model in the world with a 50% performance increase compared to previous version. So we did have the 01 preview model before in 01 mini. So now it is the full 01.

And now you have the 01 pro model as well. So there's technically kind of three different tiers of this 01 thinking model. You got the smaller one in 01 mini. You have the full 01 and now you have the 01 Pro, which is only available in that $200 per month chat GPT Pro subscription. So the new pricing strategy has led to a little scrutiny, but a lot of discussion about whether this marks a revolution and AI capabilities or a shift toward exclusivity, potentially creating an elite club for those that can't afford it. So the 01 Pro model, like I said, is only available with that $200 plan. And it promises improved performance by using more computing power to tackle complex problems in math, science, and coding.

So OpenAI's CEO, Sam Altman, and engineers started to announce these updates. So Open AI is doing this again, like us, Monday through Friday. They're doing it live at 10 a.m. Pacific time, which is noon central here. I'm in Chicago. So we're going to be covering the rest of these updates. And at the very end, I'm going to tell you what I think is coming over the next 10 days. So we already saw two of these releases. We have 10 more.

So essentially Monday through Friday of this week and Monday through Friday of next week. But hey, live stream audience, thanks for tuning in as always. But let me know what do you think is coming in these next 10 days with Open AI? So a couple other things to know about these new. updates. I did kind of an additional podcast just on this because we already had a full slate of news and information and guests for you last week. So I already covered this in-depth. So if you want to go listen or watch that, you can go do that on our website. If you want to know all the information, let me see. I think that was episode,

there we go, episode 416. But it has the higher cost, the new 01 model. And also unlimited use. And I think that is important because I think a lot of these new features that we're going to be getting over the next 10 days are either not going to be available on the standard $20 a month chat GPT plus or it will be extremely limited. All right. So more on that in a minute. And also let me know. What are your predictions, everyone? Richard says maybe unlimited tokens.

Michael says $200 a month for unlimited feature, including their latest model. So yeah, let's see what we get. Also, I love 60 minutes, right? So seeing Open AI actually on 60 minutes last night was pretty cool. But yeah, if you missed it, Open AI did kind of, again, preview or tease its live video streaming capabilities in the advanced voice mode inside of chat, GBT last night on 60 minutes. Obviously, that is not a feature that's available yet. They've been teasing it now for many months.

So we'll see if we actually get access to that. Microsoft shipping as well. Here's one I'm pretty excited about. So Microsoft has started to release Copilot Vision. So Microsoft has launched a new AI feature called Copilot Vision, which is available exclusively on its Microsoft Edge Vision. browser and it's designed to enhance the online browsing experience by understanding the full context of user activities while being able to essentially talk with a live AI agent who can

see what's on your screen. All right. So the new feature allows copilot to read along with what's on your screen and interact with users as they navigate the web, making browsing more interactive and less solitary. So copilot vision right now is currently being rolled out as a preview to a limited number of pro subscribers through copilot labs in the United States, initially working with a select set of websites as well. So users have the option to enable co-pilot vision, which acts as a second set of eyes, scanning and analyzing web pages to provide insights and assistance, such as planning activities or shopping. So privacy and security are prioritized with Vision being entirely opt-in in ensuring that all shared data is deleted after each session. So only co-pilot's responses are logged, according to Microsoft, to enhance safety systems.

So Microsoft emphasized that Vision does not store or use data from publishers to train their models, maintaining a focus on copyright, user privacy, and safety. So feedback from early testers, including third-party publishers, is being used to refine and improve vision with plans to expand access and functionality over time. So we'll probably be doing an actual dedicated episode on this vision mode pretty quickly here. And thanks to our friends from Microsoft who are getting me on the list here. So hopefully we should have a show on co-pilot vision. Maybe we'll have someone from Microsoft come on the show as well. personally, I am super, super excited for this feature.

I will see how many websites this actually works on. But I think this is a big step toward the future of work, right? And I know this sounds weird and I've been talking about this now for multiple years. I think the future of work, yeah, at least in the short time, I still think it's going to be in front of a computer. But if you're a knowledge worker, I think you're going to be orchestrating a multi-agent environment. But I think you're probably going to be doing that with your voice. right? That's the reality is, well, you can, most people can speak about two to four times

faster than they can type. And all these new AI systems, right? Advanced voice mode, which is what Microsoft uses, Gemini Live, right? They can hear, they can understand tonation changes, cadence changes in your voice, right? So when you're happy, it can kind of pick that up, right? Which is a little weird. I get it. But I do think, this new copilot vision feature as well as, you know, what was what OpenAI can again tease on 60 minutes last night with advanced voice mode with live video preview. I think this is the future of how we work, right? Think of that being on your desktop.

And yes, right now we just see it from copilot, co-pilot vision and it's just working inside of Edge, which is a great browser, by the way, it's based on Chromium. So if you're a Google Chrome fan, your transition to Microsoft Edge will be easy. breezy, everything works there. But I also think of when this rolls out to your desktop, right? Whether that's in the chat, GBT desktop app or in Windows, you know, Microsoft 365 copilot, right? But essentially, an AI agent, being able to see and understand everything that you're working

on and then being able to talk to it to help you either execute and finish tasks or if you just have questions about something that you're working on. So pretty exciting stuff from Microsoft there as well. All right. Let's get this thing. Let's keep going because there is a lot. All right. Google. My gosh, Google went wild, wild this past week. All right. So we could have done a show on just what Google kind of shipped out in AI. So we're going to do a quick bullet point. So number one, a new model. Yes, another new model. So the new Gemini experimental, 1206.

So when you look at those 1206, right, that means December 6. So it has surpassed OpenAI's ChatGPT 40, which is their 1120 version. So essentially, OpenAI in Google in late November had this back and forth kind of spat. And, you know, first, chat GPT released their 1120 version. and then Gemini released their 1121 version the day after. And the two of those models were actually going back and forth in terms of what was getting the highest scores on the chatbot arena leaderboard in terms of ELO scores, which is again, it's like a blind taste test, right?

So what Gemini did? Well, because originally their 1121, Gemini overtook chat TBT 1120, but they were actually going back and forth. So Gemini just said, nope, we're releasing a brand new version, the 1206 version of Gemini. but you can only really use that inside of their developer sandbox. So if you're still going to the front end, right, if your company is using Google Gemini, although it's very improved,

it is still using a very old model of Gemini, usually between three to nine months. And Google, unfortunately, doesn't actually tell you what version, right? So if you go inside of Google's AI studio, I think there's like a dozen different versions between their, you know, their flash models, They're on-device, Gemma models. They're big models too, right?

They're 1.5 Gemini models. So you don't really know what you're getting on the front end of Google Gemini. So you've got to go on the back end. But regardless, pretty big news from Google when they shipped out another version of their big model. We also saw a lot of hardware AI updates from Google with Android. So their latest feature drop for Android introduces enhanced AI capabilities through Gemini, allowing users to perform tasks like calling contacts, drafting messages, and controlling device settings more intuitively with a special focus on improving user experience for pixel phones.

Google got into the weather game with AI. So Google DeepMine announced GenCast, highlighting its successful outperforming of the ENS weather model in forecasting. But this was based on 2019 data, and that was published in the journal, Nature. So Google is getting in the predictive weather game, which I think is important because now as more and more, you know, different weather services start to potentially use this model. You can just have better weather prediction. So pretty big news there from Google and Gencast. And then we saw Google's new generative AI video model, VO, is now available in private preview via Google's vertex AI platform, right?

their generative AI video, text to video, photo to video. So it's offering businesses the ability to create high quality HD videos from text or image prompts with built-in safeguards to prevent harmful content in copyright violations. And it's pretty interesting there because that release actually, again, puts Google ahead of Open AI, who's competing SORA product has yet to launch, at least as of this minute. All right. And then last but not least, Google also. released its new genie 2 model, which is a world simulator. So this is a large-scale foundation world model that can generate diverse,

action-controllable 3D environments from a single image prompt, offering different perspectives and interactive objects like doors or exploding barrels. So this tool allows artists and designers to rapidly prototype complex scenes with realistic physics effects. So yeah, obviously for something like real-time video game development, that's huge, you know, video in creative fields. But you're going to be hearing a lot more about these world models in the near future. And Jeannie, too, very impressive. We cover that in the newsletter. And this is all pretty interesting here, y'all, because this is right before Google CEO Sundar Puchai just said in an interview that AI development is slowing down.

down because the low hanging fruit has been picked. That's interesting, right? It's interesting for the Google CEO to say, oh, yeah, AI development has been slowing down. Yet Google had their biggest week of AI developments and updates probably ever. So is it actually slowing down? I don't know. I don't want to argue with one of the smartest people in AI in the world. But, hey, again, if your left hand says one thing and your right hand does something completely opposite, I don't know.

I'm going to judge the actions, not the words necessarily. All right. More AI. Oh, gosh, I didn't even mention XAI and Grock in the little preview there, but we got some updates there. But one of the biggest updates is on the money round. So Elon Musk's XAI company has successfully raised another multi-billion dollar fundraising round with $6 billion according to its recent regulatory filing.

So if this sounds familiar, well, it's because they also raised a $6 billion round in May of this year that valued the company at $24 billion now. So now this latest fundraising round values XAI and over $40 billion, highlighting the significant growth potential in market impact. So yes, Elon Musk has his XAI company. So if you use Twitter or GROC, that is kind of under the XAI umbrella. So although the names of investors and investment groups were not disclosed in this latest $6 billion fundraising round, it is interesting because this one reportedly had a lot of smaller investors with some investments coming in at under $80,000. So it's pretty interesting. So according to Bloomberg News, XAI was seeking this funding at a $40 billion, sorry, a $40 billion evaluation, which does not include the newly raised capital. All right.

This is like a big tech 101, talking about all the big companies. So we go from OpenAI to Microsoft to Google to X slash Twitter straight into. Amazon. Seems like Amazon's kind of been on the sidelines a little bit. Aside from pouring billions of dollars into infropic, well, now Amazon is officially in the frontier model, large language model sphere. So Amazon Web Services announced a new family of multimodal generative AI models called Nova at its reinvent conference last week. So the Nova family includes, for text-generating models. So it has the micro-light Pro and Premiere,

with the Micro, Light, and Pro versions available immediately, and the Premier version set for release in early 2025. So right now, Amazon's new models are optimized for 15 languages and offer varying capabilities with Micro providing the fastest response time and Premier designed for complex workloads and creating custom models. So Nova's text models feature large context windows with micro handling up to 128,000 tokens and light and pro supporting 300,000 tokens. So pretty impressive there that Amazon's first frontier model already has a larger context window than the leader in this space open AI. Obviously, Google is the number one in terms of context window.

But again, that's only if you're using it on the back end as a developer with two million. token context window. But also Amazon did plan that they're looking to expand the context window to over 2 million tokens by early 2025. So what all that means without getting into too much depth, right? So chat GPT, if you're using it on the front end as an example, has a 32,000 token context window. So you essentially convert words and tokens, but that means essentially chat GPT after about 28,000 words, of back and forth, it will start to forget things, right? So when you talk about these frontier models with large context windows,

that's important because the more you work with them, you want them to be able to understand. And sometimes you might copy and paste just large, large groups of text in there. So it's really important to understand things like a knowledge cutoff, right? So how up to date is a model? And then also what is its context window? You really have to understand the rules of the game if you want to play on the court.

But that wasn't all. AWS also introduced two generative media models. So it had Nova Canvas for image generation and editing and Nova Real for creating short videos. So Real can currently generate six second videos with plan for longer, longer videos in the future. And both Canvas and Real include controls for responsible use, such as watermarking and content moderation to prevent the generation of harmful content. ADWS did remain a little vague about the data used to train these models, citing competitive

advantages and potential legal issues as reasons for limited transparency. Yeah, that's going to be the name of the game in 2025, transparency and legal issues. So looking ahead, Amazon and AWS plan to release a speech-to-speech model in quarter one of 2025 and in any to any model by mid-2020. So Amazon finally getting into the fold here. So we've had the Amazon Q platform, which is essentially a software platform that you would use, especially if your company is on AWS, Amazon Web Services, but you essentially plugged into other models. Right.

So now Amazon for the first time, hey, at least they got in before 20, 25. So they're at least a little faster than Apple, right? But finally coming to the mainstream with its Amazon Nova frontier models and immediately benchmarking not above the Google Geminize and the GPT40 from OpenAI, but at least in the same class, right? So if you're looking at kind of tiers of models instantly, even though it's not the best in the world, Amazon is going from zero to top tier. tier instantly. All right. And it should be interesting to see if this is also a way that Amazon maybe starts to rely a little less on Anthropic as well, because Amazon has invested nearly, I believe,

$10 billion into Anthropic. And they're using a lot of it on the back end inside of its Q platform and in other places. So in the same way that Microsoft is starting to develop a lot of its own internal models, which are pretty impressive, to maybe. reduce its reliance on open AIs models. We're seeing the same thing here with Amazon. 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.adobie.com. This is a lot, y'all. Let's keep going. Let's keep going.

11 labs. Yeah, that text to speech platform, that's actually really good. Well, they've just launched a new conversational AI agent platform. designed to enable the creation of customizable and interactive voice agents. So the platform is noteworthy and known for its ability to integrate voice capabilities into the web, mobile, or now telephone applications in just minutes, potentially transforming how businesses interact with customers. So a key feature of this new platform is its ability to handle turn-taking and interruptions, utilizing a real-time model to predict when a speaker has finished talking,

which is particularly valuable in corporate settings. So the new platform from 11 Labs supports 31 different languages, aiming to communicate with customers in their native language. 11 Labs also highlights customer support as a primary use case with the AI capable of handling customer inquiries around the clock, reducing wait times, and improving customer satisfaction. The conversational AI can also be used for outbound sales, scheduling, interactive gaming characters, tutoring, and more,

offering versatility across various industries. Integration features include native Twilio supports for texting, call handling, and both server side and client side tool calling for flexibility. The technology can also connect to a variety of large language models, including Claude, GPT in Gemini models, or a custom large language model providing users with choice and adaptability. So essentially, you choose a model, right?

So you don't have to leave that up to 11 labs, which is primarily text to speech. And you can also upload your own documents. So again, this is a brand new model. We haven't really gotten a chance to test it out. But again, we are talking about a drag and drop, AI voice agent that can handle your calls, can handle your inquiries, you can put it on your website, you can upload your own knowledge base, you can work with some safeguards, you can use

it for outbound call scheduling. I mean, I've been saying this for a while. Agents in voice is going to be a big focus of 2025. And it's going to be weird at first. I get it. Because does anyone actually want to talk to a robot? No. But guess what?

I also don't want to wait on hold for 45 minutes and then talk to someone in a crowded call center who I can barely hear. So it's like, is there pros and cons to each? Absolutely. Would I prefer a human? Yes. But this might be the future of customer service, right? Talking to a quote unquote live AI-H

that is powered by a large language model that is neural, right? So a low latency in response time. So this isn't like talking to Siri or Alexa, which is a terrible experience, right? So don't think of that. Think of like advanced voice mode, right? And how it feels real. It feels natural.

I think Gemini Live from Google is pretty good in this as well. A copilot uses Open AI's technology, maybe lagging a little behind, but think of all of these neural, live and interactive voice models that can kind of, you know, pick up on the tone of your voice. Like I said before, you can interrupt them, right, which I know seems rude. And everyone's like, why would you interrupt a, well, I don't know, when I'm on the phone, I sometimes interrupt someone as well, right? If an agent, if a customer service agent got something incorrect and they're about to give

me a three minute spiel on something's wrong, I'm going to try to nicely interrupt them, right? But the problem with talking to, quote, unquote, you know, smart AI assistance over the last decade is they're, well, dumb. And they don't have that conversation, like the ability to kind of carry on knowledge from query to query, right? If you ask Siri or Alexa, one thing and then do a follow-up question, they have no clue what you're talking about, right? So this is different.

And then being able to, you know, for these companies to upload their own documents, to kind of build their own guardrails, and then to actually have access to all that information to hopefully improve their products and services and customer experience in the future, I think is pretty big. All right, we got more. Here we go. So president-elect here in the U.S., Donald Trump,

has appointed David Sachs to be the new AI czar. So a move that could significantly actually impact the broader AI. industry and open AI. So David Sacks, if you haven't heard of him, he's a co-host of a popular AI podcast called All In. It's more of a tech podcast, actually, not an AI podcast. Actually, it's not AI at all. They talk AI, but it's a very popular tech podcast as well as he is the former CEO of PayPal and a pretty prominent venture capitalist. So this appointment alongside Elon Musk taking over an official role in the U.S. government in the leading of the Department of Government efficiency.

So it introduces some influential figures who have been critical of Open AI. So David Sacks, whose VC firm Kraft Ventures, has invested in Musk's company, XAI, has criticized Open AIs transition from a nonprofit to a for-profit entity. So Musk, like we said, was an original co-founder of Open AI. who left after some internal conflicts and then launched XAI and has been pretty vocal regarding open AIs direction. So now you have two people with official government titles who are kind of, you could say, pro, you know, acts or pro Twitter, pro grok and anti-open AI.

And that's interesting, y'all, because you have to look at the broader implications. Because we're going to talk here just in a second about some negotiations that are going on between Microsoft and OpenAI. But whether you know it or not, Open AI is going to power the majority of our online interactions in 2025. All right. So Microsoft uses OpenAI's GPT40 technology. Apple, you know, essentially for the new Apple intelligence that's rolling out, the majority of, more complex queries are just passed on to Open AI. So think about that.

If you use any Microsoft Windows products, if you use ChatGPT, OpenAI, if you use any Apple products, you're essentially using OpenAI. So it is going to be interesting when you have some seemingly very vocal critics of Open AI, but also who have their money against Open AI in official government positions. This is kind of unprecedented, right? This bringing in kind of some politics into the AI, open AI scene. So right now, government regulation is a growing concern for AI development at large, with Musk and Sachs' position to shape future policies significantly.

So Musk has expressed worries about AGI misuse and supported measures to regulate powerful AI models, whereas SACS has advocated for unregulated AI development markets, suggesting some potential changes or challenges to current regulations, including current president Biden's AI executive order. This is interesting, y'all, because I don't, let me be honest, I'm not a huge fan of this. I'm not a huge fan of this. Like, I like that there is, you know, an AI SAR, right? Cool.

But I would have liked to have someone with some AI research experience. So, yes, David Sacks is a very prominent figure in the tech scene, one of the most well-known people in venture capital. But he's not really an AI person. He launched a little AI company this year. But, I mean, he's a tech person. He's a VC person. Yes, he was the CEO-O PayPal. But I would have liked to have someone in this position who has,

some actual AI research experience, right? And that's not a slight at him or his credentials, right? But I think when you want to elevate someone to a high government position, you would hope that that person is among the most qualified in the world for that specific positions. If this was the czar of technology, it's like, okay, sure, I can see that, but it's not. So pretty interesting here. And I think that we're going to have to keep an eye.

eye on that. Yes, and thank you, Michael. Also, it is a kind of a dual position. So also, he is the crypto and AI czar as well. All right. Speaking of that Microsoft in OpenAI relationship. So right now, OpenAI is in discussions to remove a clause from its agreement partnership with Microsoft that would exclude Microsoft from accessing its most advanced AI models.

Once artificial general intelligence is achieved. So that is according to a report from the financial times. Let's talk about what this means. It's actually pretty big. So there is a current provision that would be void or sorry, which would avoid Microsoft's access to AGI. And it was intended to prevent the misuse of powerful technology for commercial purposes, with ownership defaulting to Open AIs nonprofit board.

So as Open AI restructures to become a public benefit corporation away from a nonprofit organization, the change could encourage further investment from Microsoft, which has already invested reportedly over $13 billion into Open AI, which is crucial for funding the costly development of advanced AI models. So OpenAI CEO Sam Altman has highlighted the need for the restructuring, acknowledging the unforeseen scale of capital required in the shift from a non-profits research lab to a for-profit product company. So the discussions include redefining AGI as a continuous process rather than a single point with societal input playing a role in its definition. Open AI plans to retain an

independent nonprofit entity focusing on its mission to benefit humanity while ensuring, it receives value from its stake in the new for-profit structure. The restructuring has faced criticism. Like we said, notably from Elon Musk, who was an early co-founder of OpenAI and then broke away from the company and is now suing them in part because of their shifting away from a nonprofit to a for-profit company. Also, this clause is huge. I'm not going to spend a ton of time because I actually did like an hour-long show.

about six months ago on this. But essentially this clause kind of restricts or prohibits Microsoft from using any model that you can say, oh, yeah, this new model from Open AI, this is AGI. Right. So artificial general intelligence is essentially when a single AI model is smarter than most all humans at most all tasks, right? And we've talked about it before on the show here as well. who defines AGI, artificial general intelligence, right? Right now, no one, there's no one official definition.

It is a moving goalpost. The goalposts are moving here, right? And again, I've said this. If you look at definitions from 15 years ago, we've achieved artificial general intelligence, right? But the conversation and the narrative and the definition keeps changing. So according to the original clause, essentially, Open AI's board is the one that says, yo, yeah, we've achieved AGI.

And then at that point, Microsoft's access to certain technology is a little restricted. So some of these ongoing negotiations between OpenAI and Microsoft are actually extremely important because the overwhelming majority of businesses out there, especially in the U.S., are running Microsoft 365 and probably Microsoft 365 co-pilot. So if you want your company, your enterprise to have access to the most powerful AI that's available right now, you have to have it, right? But there is this, oh, this tipping point of AGI because if Open AI were to say, yes, we've achieved AGI, then things get complicated in terms of what Microsoft can and can't

have access to. So I know it's this ongoing discussion about this small little clause, but it's a lot of important to pay attention to. All right. I didn't forget about meta. Because they quietly came in and released Meta Lama 3.3, a new multilingual, large language model that is benchmarking way above its weight. So Lama's 3.3, the most powerful version of it right now, is the 70 billion parameter

model. And here's the thing, y'all, from a benchmark's perspective, this 70. billion dollar or 70 billion dollar 70 billion parameter version of llama 3.3 is already comparable to the 405 billion parameter version of llama 3.1 but at a significantly reduced cost in computational demand that's wild. So essentially a model in 3.3, the 70B version, a fraction of the size is already outperforming uh, Lama 3.3. So essentially a model, uh, a model in 3.3, the 70B version, a fraction of the size is already outperforming, uh, Lama 3.1 405B. So the new model is available under the Lama 3.3 community license agreement, allowing free use with attribution except for organizations with over 700 million monthly active users who need a commercial license.

So according to meta benchmarks, Lama 3.3 outperforms similarly sized models, including Amazon's new NOVA Pro in multilingual dialogue and reasoning task. though Nova excels in human e-val coding tests. So Lama 3.3, according to meta, achieves a 91.9, sorry, 91.1 acubricy rate in multilingual reasoning tasks. Also, the cost for token generation is bonkers. It is so cheap. It is way cheaper than other frontier models like OpenAI's GPT40 or Anthropics Clawed Sonnet 3.5 new. It is a fraction of the cost.

Right now, token generation for meta 3.3 is a cent per million tokens. Like, that's so cheap. If you would have said this like a year or two ago, you said, yeah, right. That's 100 years away. That's 50 years. Whatever. It's here already. So meta size emphasized, which I like.

So meta emphasized environmental responsibility. They said they achieved a net zero emissions for the training. phase through renewable energy, right? Everyone's always like, oh, large language models, energy demand, energy consumption in the training, right? Well, there you go. Meta just said they achieved net zero emissions for training the new Meta Lama 3.3. So this medal includes advanced features such as a 128K token context window and what

they're calling grouped query attention or GQA for improved performance. So right now, Lama 3.3 is available for download on platforms. like Meta, Hugging Face, and GitHub with resources for safe deployments, such as LamaGard 3 and Prom Guard. So very impressive here, right? They literally snuck this in, right? I think it was like Thursday or Friday, just toward the end of the week. After all this noise, right, like we said, Open AI, Microsoft, Google, Amazon,

and meta just sneaks in. You know, it's not truly an open source model, but it's, it's, it is, it is, it is, open source in some ways. So they slip in a free model that everyone can download and fork and build off of, right? That's the big difference. All these other models are proprietary. You can't download them and modify them, right? You can still fine-tune them and, you know, add in rag with your data,

but you can't literally download them and fork them, right? You can with meta-slamma. And the fact that it is this cheap, meta is a serious, serious contender. Right. I think, you know, sometimes you always think of, you know, Open AI and Anthropic and Google as like a 1A and meta as a 1B. Nope. Nope. Meta is definitely 1A because they are pushing. They are pushing the prices down. They're pushing the power and the benchmarks up. And then the fact that it is somewhat open source changes what these proprietary models actually have to do to keep up. All right. That's essentially it. But I wanted to give you guys some. thoughts on the Open AIs, 12 days of ship miss.

What do we have coming? 10 more days of announcements. All right. So Joe said, I would love to see an API playground, you know, June here from YouTube asking about operator. All right. So let me just quickly tell you this 12 days, which I think is smart, because Google's been releasing a ton and they're about to release a ton more.

they didn't really market it. For a startup, Open AI is great at marketing. They're going to market strategy. Might leave a little to be desired, right? We get a lot of wait lists. We get a lot of like blog posts when there's new features. So what we've already seen, we've already talked about it. We saw that new, the full version of 01.

There is the new chat GPT Pro, the $200 a month subscription that includes 01 Pro. We also saw OpenAI's reinforcement fine-tuning research program. All right. So those were the announcement. so far. But then what are we going to get in the next 10 days? Because I think what Open AI releases here is going to set the tone for 2025. So you need to pay attention. And we're obviously going to be covering it every single day in our newsletter and probably on the show as

well. So what's next? Well, we've seen in the past couple of days what looks like a new version of SORA, which is Open AI's text to video model, image to video model. And as well as a leak of this last week, which we covered. I think we're going to see it. But who's going to get access? I don't know. Personally, I think if you're on the free plan, you're not going to get access. I don't think to anything. If you're on the $20 a month plan, you might get a little SORA if they do actually release it. I do think that they're going to release it in a very, in a very limited fashion. But I actually think for the most part, you're really only going to be able to get real utility out of it if you're on that $200 a month plan, right, the chat GPT pro plan. I think

if the chat GPT plus $20 a month plan gets SORA, it's going to be extremely limited, right? Extremely. So we'll see about that. So the other things that I think are coming, so over the next 10 days. So we do have SORA. I do think there is going to be a huge differentiation on this $200 a month pro tier.

So I think we're going to see some advanced voice mode updates. And many things like that are going to be for the pro mode only. So I think advanced voice mode is going to get tools. I don't know how many tools they're going to get, but right now, advanced voice mode is not super usable because if you go in and type with advanced voice mode, you can no longer use that chat. It can't see anything, right?

So this demo we saw in 60 minutes, no one has that yet. You don't have this live ability to show it video and it can see and respond. I do think we're going to get some updated tools inside of advanced voice mode. whether that is the new live video, whether it's it being able to see and understand things on your desktop, similar to the new co-pilot vision, being able to upload files aside from, you know,

the new 01 model can now upload PNGs and JPEGs. So I hope that the new advanced voice mode also gets access to real-time information on the internet. So I do think we're going to see some advanced voice mode updates and features. Also, some of these things are rumors and rumblings. Some are a little more confirmed, right? So I think we're going to see projects pretty soon. So projects is a way that we'll be able to organize chats. So think of folders, right?

And being able to organize your chats in folders, your files, and custom instructions in projects. So that one, I would say is a little more than a rumor. A couple people have been sharing about that online with some screenshots. I do think we're going to see API costs go down. Again, whether that's 01, the new 01 models or the GPT40 model, I think we're going to see both an update to the GPT4O model. And it could just be in cost on the API side. But I do think it is very likely we might see a GPT 4.5. We're not going to get a, I don't think we're going to get a GPT5, the Orion model.

I don't think so. There's no need to. Open AI's 40 model, depending on what benchmarks you look at is still one of the most powerful models in the world. So I think we just may see a 4.5 update. So similarly, how we saw a GPD3, GPD3, 55, GPD4, I think we're probably going to see a 4-5 update. But I think a lot of these updates are going to be coming at the end of this kind of 12 days of open AI. And then I do think we're going to see an agent's preview. So what was co-named operator, I do think that we are going to at least see a preview.

I don't think that is going to become available. I think there might be a wait list and a blog post, but I think this might go into waitlist and official announcement because we haven't seen anything official from OpenAI with agents, again, codenamed operator. We haven't seen anything official. So I think they're probably going to get finally official with showcasing operator and probably a waitlist, but not a release, I don't think.

And also with the new SORA updates, which we'll probably see, that also will likely mean improved image generation because we know the SORA model has the ability to generate photos as well. All right. That was it ton, y'all. Thank you for tuning in as a quick recap. All the AI news that matters. So Open AI has unveiled a new, new models and pricing strategy with a $200 a month, chat GPT Pro plan. Microsoft introducing and rolling out to co-pilot vision. Google released just about every single AI update under the sun.

New Gemini model, Android for the pixel, weather models, AI video, world simulators. Then Elon Musk Company XAI secured another $6 billion funding round. We saw Amazon release a ton of new models with its Amazon Nova. 11 Labs gets into the conversational agent game with its custom. voice agents. David Sacks was appointed as AI and CryptoZar by incoming president-elect Donald Trump. Open AI is considering changing its AGI clause to maintain and improve its Microsoft partnership. Meta unleashed Lama 3.3. And we just gave you a couple predictions on what we're going to see from Open AI over the next 10 days. I hope this was helpful. If it was, please share this.

post this, don't keep all this good information to yourself. Make sure to join me tomorrow. The show link is already posted. Get your questions in. I want you to put me on the hot seat. I also want you to go to your everyday AI.com. Sign it for the free daily newsletter. Thanks for tuning in y'all. 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. Stand control with the ability to step in and refine at any time. See it today at firefly.adop.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. a little more AI magic. Visit your everyday AI.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.