Everyday AI Podcast – An AI and ChatGPT Podcast - EP 468: AI News That Matters - February 24, 2025

Episode Date: February 24, 2025

Grok 3 is already swirling in controversy. Google released their jaw-dropping Veo 2 AI video model, but not in the way you'd think. And Microsoft quietly unveiled a new piece of hardware so big, it could change every aspect of our lives. Not just the AI stuff. Join us as we bring you The 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. Grok 3 Controversies2. AI Censorship and Political Stances3. OpenAI and Chinese AI Misuse4. Advancements in Robotics5. Google's AI Co-Scientist System6. Launch of Mira Murati's New AI Startup7. Google's VEO 2 AI Video Model8. Microsoft's Quantum LeapTimestamps:00:00 XAI's GROC 3 Launch Controversy03:56 GROC Three: Advanced AI and Modes07:27 Disputing AI Benchmark Claims10:49 Grok Chatbot Censorship Controversy16:59 Helix: New AI Model for Robots18:20 Helix Robots Store Groceries Autonomously21:23 AI Coscientist: Revolutionizing Idea Generation26:04 AI Cuts and New Startup Reveal28:12 "Thinking Machines Lab's AI Ambitions"33:32 "V-O Two Model Now Accessible"37:32 AI's Media Impact and Quantum Leap41:15 "Quantum Computing Milestone with Majorana"43:33 Anthropic's Claude 3.7 SONNET RumoredKeywords:Elon Musk, Grok Three, AI controversies, AI benchmarks, synthetic data, XAI, GROC team, deep search, benchmarking, synthetic data, quantum computing, quantum leap, AI misuse, OpenAI, Anthropic, AI co scientists, robotics startup, humanoid robots, Microsoft, Majorana one chip, quantum processing, AI video model, Google's V O Two, video generation, Google's AI coscientist, AI safety, US AI Safety Institute, Mira Mirati, Thinking Machines Lab, Figure robotics, Helix model.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 and 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. Elon Musk's grok three is already facing multiple controversies.

Google's VO2 state-of-the-art AI video model has been released, but not in the way that you would think. And Microsoft is going quantum and apparently discovered a new type of matter. I don't know. Like, it's Monday. And it already feels like we're at the end. end of the week in terms of the amount of AI news that is just hitting us. And that's not all. We might also see a new update, a new model from Anthropic, from Claude after, I don't know,

waiting like seems like 30 years. All right, we're going to be covering those stories and a whole lot more in today's AI News That Matters. So welcome to Everyday AI. What's going on, y'all? My name's Jordan Wilson, and I'm the host of this show. And this is for you. So we do this every single weekday, Monday through Friday, bringing you the real updates that matters. So on Mondays, we do our AI news that matters, but we do the everyday AI show every single weekday. So if you're looking to grow your company or grow your career by understanding the latest in Gen AI, you are in the right place. The other right place for you to be, your BFF is going to be our website. That's your everyday AI.com.

You can go there. You can, number one, sign up for our free daily newsletter where every single day we recap not just our podcast, which is usually bringing you some exclusive insights with interviews and everything like that. But on there, you can go listen to like now 450 episodes, all sorted by category from the world's leading experts all for free. So make sure you not only go sign up for that free daily newsletter where we will be recapping today's show. But also go check out, I don't know, one, two, 20 episodes and be there. the smartest person in AI at your company. That's what we are here in trying to get you to do.

All right. And also make sure go listen to episodes 443 through 447. Some people are asking recently like, hey, Jordan, where do I start? This looks intimidating. You have like 500 episodes. Go listen to those. That's our 2025 AI predictions and roadmap series. So if you have not listened to episodes 443 to 447, please go do that.

All right. Enough chat, chat, y'all. Let's get into the AI news that. for the week of February 24th. And also, thanks for joining us live. If you're normally on the podcast, come hang out. It's fun. Michael and Michelle, Sandra, Philip, a lot of people joining on the YouTube machine this morning.

Love to see it. Samuel, Bethany, Lauren, Jackie, big bogey face, Alan, Natalie. Too many to name this morning. Thanks for, thanks for tuning in, y'all. All right, let's get into it. Let's start at the big story, what everyone is talking about. Elon Musk's startup XAI has launched its latest AI model GROC 3, claiming that it surpasses competitors like OpenAI, Claude Anthropic, and Google and others.

So GROC 3 is said to be significantly more capable than its predecessor, GROC 2, and is available to premium users of the social media platform axon. It's rolled out to free users as well. It's actually been a confusing rollout from the Google. GROC team because when it first rolled out, you had to have a premium plus and then you had to have a premium and then they doubled the price of premium and then they made it, you know, somewhat available to free users, but with not a lot of, you know, communication. But it is available.

And the model has been tested on standardized exams and everything from math, science, and coding, reportedly outperforming existing AI models. More on that reportedly here in a bit. So Musk described GROC 3 as scary smart with enhanced reasoning abilities and noted that it is trained on synthetic data, allowing it to reflect on its mistakes and improve logistical consistency. So the launch of GROC 3 also includes a new product called Deep Search. All right. You know, Grock had to go against the grain and not name itself deep research like the other deep research tools from Google, perplexity and Open AI. So you have the deep search mode in GROC3, the new voice mode and people are going wild about the unhinged mode.

So I don't know if you're 13 and, you know, giggle at profanities. Maybe it's for you. And as well as a kind of a think deeper mode. So similar how you can kind of control certain models to use a little more compute or to use a reasoning model. You have that with the new GROC 3 as well. So the model's rollout is part of the beta phase with ongoing improvements expected and the voice assistant features slowly rolling out already. I do believe the latest. I haven't used the new voice features, but I did see it was released to many users as of late this weekend.

So XAI has expanded its GPU cluster, doubling the size to support the training of GROC 3. So originally it was reported that they had 100,000 of Nvidia's most powerful GPU chips. Well, it turns out they actually reported they have 200,000 of Nvidia's most powerful GPU. So they really just willed Groch 3 into existence. So you have to tip your cap to the XAI team, right? Going from essentially zero to Groch 3 in like a little more than a year. and the GROC, the release of GROC 3 comes amidst intense competition in the AI market with previous models from OpenAI and Google setting high benchmarks that GROC 3 has now either topped or has at least caught up to in almost all benchmarks. So, you know, I'm curious for our live stream audience if you are using GROC, right?

Michael says, hey, I'm 35 and giggle at the profanities. So, you know, let me know if you're actually using grok. If you want to hear more about it, I don't know. It's going to be grabbing a lot of headlines, I think, the first couple of weeks. And speaking of that, yeah, we actually have two more Grock stories to start off with. And they're pretty significant that I think we kind of have to talk about here. So speaking of those benchmarks, there is a controversy that has erupted online between Open AI and XAI concerning those exact benchmarks.

So Open AI accused XAI of presenting misleading benchmark results, which XAI co-founder Igor Babushkin defended highlighting a broader debate about transparency and AI performance reporting. So here's the gist, and I'll try to break it down hopefully simply. So XAI published a graph showing GROC 3 outperformed OpenAIs O3 Mini High model on the aim. That's the AIME 2025 math benchmark, a test whose validity as an AI benchmark has by question by some experts. So the dispute centers around the omission of OpenAIs O3 Mini High Score at Consensus 64. So consignes 64 or Consensus 64 is a method that evolution. allows a model 64 attempts to answer each question, potentially inflating performance scores.

But when measured at 1, which is the first attempt, GROC 3's performance was lower than Open AIs 03 mini high, contrary to XAI's claim of GROC 3 being the world's smartest AI. So essentially, if you gave GROC 1 attempt, it did not beat. Open AIs 03 Mini and some other models, and they use this O, or sorry, this cons, 64, which essentially gives a model 64 attempts to see if it can get the right answer. So parts of this, I understand, right, because large language models are generative. They're not deterministic. So they are like a controlled roll of the dice, right?

So you could do the same prompt 10 times, get nine different answers. You can get one different answer. depends on what you're asking, and if there is a definitive answer or not. So I personally, though, think it's worth noting this whole cons-64 method, because, you know, it's essentially now what he said, she said. Open AI said, nah, Grock, you guys lied on these benchmarks. And Grock's like, no, you use cons 64 as well on your benchmarks. Here's the thing that I didn't really see anyone else talking about.

So, yes, OpenAI use cons 64 or the best of 64, but only when comparing its own models, right? And kind of going to show that when you gave a model like O3 Mini high extra compute, that is what makes it the high variation of that, right? They were essentially showing how much extra juice you can squeeze out of the model, you know, on this high mode or giving it 64 attempts. to get an answer right. But I haven't seen anyone except now GROC in XAI, use this cons-64 benchmark to compare itself to different models.

That is not something that Open AI did, right? In this instance that they're talking about, Open AI was just comparing its own internal models with the cons-64, not external models. So a bunch of geeky benchmark drama going on in the AI world. But that is not the only controversy that Grock is already facing. Yeah. I would love to say I'm surprised, but I'm not.

All right. And y'all, I'm not trying to get things political here. I know people are going to, you know, people sending me hate mail. This is just what happened. All right. I don't know why people get so mad like at me. And they're like, oh, Jordan, you're making this so political. I'm like, no, this is just facts.

This is just facts. This is what happens. So save your grumbling for someone else. All right. So XAI's Grock 3 chatbot has been reported to have its search capabilities censored, sparking controversy over free speech and the truth-seeking promises made by Elon Musk. So this is according to reports and recent users. And when I saw this, I did replicate this.

And yes, this is actually or it was actually true. So again, Igor Babushkin from XAI confirmed a. I confirmed a system prompt update was reverse after user feedback indicating internal misalignment with company values. So users noted that GROC 3's search instructions had previously told the model to disregard anything that it saw online, right? Because GROC has access to the internet and obviously has access to Twitter. So there was a system instruction that told GROC to ignore anything.

that said Elon Musk or Donald Trump spread misinformation. Right? So a couple online sleuths, you know, it's whenever new models come out, people try, you know, thousands of different things to see, you know, what's maybe not correct, what's maybe misaligned in a model. And this is something that was obviously pretty bad, right? If you're trying to be the free speech platform and you are giving specific system instructions saying, hey, if anyone asks about misinformation and who spreads it, don't mention Elon Musk or Donald Trump, right? Not at all.

So the move, like I said, contradicts Musk's previous assertions that X, the platform hosting GROC is dedicated to free speech. So GROC 3 has taken unexpected political stances, also listing Donald Trump first when asked about who deserved a death penalty. Yeah, that's not good. and labeling, like we said, Musk as a major misinformation source. XAI has responded by adding a prompt to prevent GROC from commenting on death penalty cases. Yeah, so people were actually saying, hey, who deserves the death penalty? And it came back naming names, including U.S. President Donald Trump. So that's not good.

So the team has said that there is a permanent fix. And since I'm not putting those type of prompts out there into. existence, but since people have shown that that has been kind of fixed. So this behavior highlights the challenges Musk face in balancing free speech with controlling narratives, especially as Grock 3 continues to express views contrary to Musk's apparent political preferences. And that's just the beginning of it. There's a whole other very serious issues with Grock.

It was readily giving out instructions. on how to make drugs, how to make chemical weapons of mass destruction, things that large language models should not be spitting out. So I guess we'll have to keep an eye and hope Grock becomes a little less unhinged because that's not good. All right. Moving, moving on. Moving on.

Now we have more AI misuse. Who would have thought? All right. So Open AI has identified and disreaching. attempts to misuse its AI tools in Chinese influence campaigns, including spreading Spanish language anti-U.S. disinformation. So according to reports, this is, so according to OpenAI, one of these campaigns was called sponsored discontent, and it used ChatGBT to generate anti-American Spanish language articles and English language comments. So,

these articles were distributed across various Latin American news sites, sometimes as sponsored contents, while comments appeared on platforms like X. Yeah. You know, all the time when I say, you know, people ask me like, hey, are you going to use GROC? Should businesses use GROC? And I'm like, absolutely not. And this is one of the reasons why I told you a couple of them and some of the controversies that GROC is facing right now. But the other thing is GROC relies heavily on X. And it has been widely shown that the X formerly known as Twitter platform not only has the highest rate of disinformation and misinformation versus other social media platforms. No social media platforms are perfect. Right. And meta also uses its platforms and its training data for

its Lama models, but not at the level in the frequency that GROC does and that GROC uses X, you know, X content. So this is just another reason why you got to be careful. And it's, there's a lot of bots, right? There's a lot of bots. So, you know, Open AI found that it's, uh, GPT technology was being used to spread, uh, you know, kind of, uh, this pro Chinese and anti-American, uh, disinformation on platforms like X. I already said that a couple weeks ago. I believe I said that in my, uh, AI predictions and roadmap series, uh, right? That, AI was going to be used in a lot of bad ways by China. So Ben Nemo from OpenAI's intelligence team noted this is the first known instance of

Chinese influence operations targeting Latin America with translated articles. So there was another campaign called peer review and it involved using chat to create marketing materials for a tool allegedly used to report protests to Chinese security services. So OpenAI has banned the involved accounts, citing violations of policies against using AI for unauthorized surveillance. All right. Let's get to robotics. So figure, a robotics startup from Silicon Valley has introduced Helix, a new AI model for humanoid robots. So the Helix model allows robots to handle objects, collaborate with other robots and control their upper bodies more smoothly.

So this new launch of the Helix platform follows Figures decision to end its collaboration with OpenAI and raise $1.5 billion. So yeah, Figure made some headlines about three or four weeks ago when they said, yeah, we're ending our partnership with Open AI because in their first Figure 01 model, it was using chat GPTs or OpenAI models, both their vision, their speech to tax, and just their full 4-0 model was using it for the figure 0-1. So now we saw this kind of what figure decided to do instead,

and it did look fairly impressive, although the demos were extremely limiting, but Helix could represent a significant step toward integrating humanoid robots into everyday home environments. So unlike, previous models, Helix does not require extensive training on specific tasks, enabling interaction with unfamiliar objects. So in the demonstration video, Helix powered robots successfully put away groceries,

showcasing their practical applications. So Figure did say in this demo video, essentially it was giving them groceries to put away in a simulated kitchen. There were two of these new figure robots, presumably running the Helix kind of model or the Helix system. And it was said that they hadn't been trained on these items, right? So there were some grocery items and some other things that were placed on a counter. And then these two figure AI robots silently put away the groceries and worked with each other. So they handed each other things, which I thought was both pretty cool, right? Because there's times like, I got home late last night.

I didn't want to put away the groceries, all right? But I had to. What I want to, humanoid robots putting away the groceries? I don't know, maybe. Maybe in the future, but a little weird. I actually found it a little unsettling that the robots weren't talking, right? Like, I think figure was trying to show like, oh, look how, you know, smart these AI, these humanoid robots are. They can, you know, first, they haven't been trained on these models like, oh, the apple goes in the dish and, you know, the cold thing goes in the fridge, right? So, uh, figure, said that they weren't trained specifically on all of these items, but they weren't communicating, which I don't know.

At least when I think about it, when and if there's a humanoid uprising, I would like for them to be talking to each other so at least I can understand what's going on. So although it shows some pretty emergent capabilities, I don't know. Personally, I found it a little disturbing that they were collaborating silently. So Figures founder Brett Adcock highlighted a major breakthrough in robot AI, like I said, achieving this Helix model entirely in-house. So the company is reportedly in talks to raise another $1.5 billion, valuing it at nearly $40 billion. So, yikes?

I don't know. Are you guys excited about this? Angie from LinkedIn says, Silent Grocery. agents are creepy and asking, are they just listening and watching us? I don't know. Joe says, I don't want chatty robots putting away groceries away when I'm trying to sleep. So I don't know. Maybe people want multiple humanoid AIs, you know, silently going around and doing things. I don't know. I would prefer hearing actually what they're saying. But that's just me. That's me. All right. Some other big, big AI news, not necessarily on the large language model side,

but Google has introduced an AI system called AI co-scientists, designed to help scientists formulate new hypothesis, potentially accelerating scientific and medical research. So this AI system uses unique methods involving AI agents that generate, debate, and refine ideas before presenting them to human scientists. So, unlike, Other AI models, AI co-scientist, produces new ideas rather than summarizing existing ones, setting it apart from reasoning models like OpenAIs 03. So the system has been powered by Google's Gemini 2.0, but can work with any large language model, offering flexibility across various fields.

So the AI co-scientist comprises several specialized agents that work together to generate, review, and refine medical hypothesis, simulating a team of research assistance. So during testing, the AI co-scientists successfully generated a hypothesis about antibiotic resistance that matched findings from an independent study, demonstrating its potential effectiveness.

So while the AI co-scientists can perform complex tasks, it is designed to collaborate with human scientists, not replace them. So the US and UK have initiated fellowships to study AI's impact on scientific research with funding from the Alfred P. Sloan Foundation and the UK government. So according to researchers, these advancements in AI could significantly augment biomedical and scientific discovery, ushering in a new era of AI empowered scientists. So I've been saying this for many years, right? I've always said the future of large language

model is many small language models, right? And we're kind of seeing that here. here with Google's co-scientists, instead of just, you know, having 10 different versions of Gemini 2.0, and each of them just take a small section of the tasks. These are specially developed, essentially small agents, and these agents just have one role. So it's thinking of it as a narrow focus. You know, sometimes we talk about, you know, ANI, artificial narrow intelligence versus this AGI artificial general intelligence and right, these large language models are trying to tackle artificial general intelligence, which means they're trying to be the best at everything. But, you know, and we're not just talking from an LLM perspective.

We're talking from an agentic perspective as well, right? That's why I think early on, the early buzz of agents in, you know, late 2022, early 2023, at least LLM powered agents. It was these agents. It was like one agent that was trying to do absolutely everything. where I think what we just saw here from Google with co-scientists is going to be the larger trend throughout the industry is you're just going to have at first maybe a handful, but then dozens, but eventually hundreds of agentic AIs working with each other, but really just fine-tuned

for one very specific task. Why? Well, narrow intelligence is much easier to achieve than general intelligence, right? So if you train a. model, or if you fine tune a model specifically on one area, let's say medical or, you know, researching potential reasons, antibiotics fail as an example, right? If you train one model on just that, it is going to perform much better than the larger model that it was distilled from that was not trained specifically on that. So pretty exciting stuff there from Google. All right. Our next piece of AI news, some unsettling one. But the Trump administration is planning to cut almost 500 roles at the U.S. Safety Institute housed within the National Institute of Standards and Technology

or NIST. So this is according to reporting from Axios. So these cuts are significant as they may deeply impact AI safety and regulation efforts, potentially leaving the AISI as gutted. So the AISI has been instrumental in overseeing AI model testing and collaboration on regulation efforts with companies like Anthropic and Open AI. So yes, for the last year or so, the big AI labs have been working with the USAI Safety Institute to make sure new models that they release are safe. well, might not be happening anymore with this new, with this recently created USAI Safety Institute, now reportedly being gutted. So the cuts also affect semiconductor production, including, quote unquote, from Axios, 74 postdocs, 57% of CHIP's staff focused on incentives, and 67% of CHIP's staff focused on R&D.

So the decision seems contradictory to the Trump administration's goal of achieving AI dominance over China, especially considering the national security implications of the CHIPS initiative. So the anticipated firings follow the exclusion of AISI staff from the recent AI Action Summit in Paris and the resignation of AISI Director Elizabeth Kelly reportedly due to political pressure. So this development is part of Trump's broader AI agenda, which prioritizes AI dominance over safety and regulation. All right. New news, I guess we have a name now from Mira Muradi's newest company. So Mira Muradi, the former CTO.

the chief former chief technology officer of OpenA.I. Has officially unveiled the name of her new AI startup called Thinking Machine Lab, aiming to address significant gaps in advanced AI systems. Sorry, Thinking Machines Lab. So the startup's mission is to make AI systems more understandable, customizable, and generally capable, according to a blog post shared on Tuesday. So it's not just her at the top of the ticket. There's some other big names from OpenAI and other big tech companies.

So John Shulman was a co-founder of OpenAI, will join Marathi as the chief scientist after he left Anthropic. Yeah, so Shulman went Open AI, Anthropic, now going over to Thinking Machines Lab. Also, Barrett Zoff, previously Open AI's vice president of research. will serve as the chief technology officer, and at least seven former Open AI staff have joined the team. The team also includes researchers from top AI companies like Meta, Google DeepMind, Character AI, and Mistral. So Marotti left Open AI in September to explore new opportunities and was reportedly in talks to raise over $100 million for her new startup. So, well, what is it all about? venture is significant for the AI industry as it highlights a trend of key talent moving to new

projects and potentially shaping the future direction of AI research and applications. So it should be interesting to see what Marotti and her teammates cook up at Thinking Machines Lab. So we don't have a ton of new information about what they're going to be working on aside from addressing the gaps in advanced AI systems and to make AI systems more understandable, customizable, and generally capable. So not exactly sure what that is actually going to mean yet, but Maradi did assemble a pretty impressive roster of talent to get Thinking Machines Lab off the ground.

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. Michael says he's excited for Mira's company. Yeah, I'm excited to see what they cook up as well. All right.

Speaking of cooking up, you can now cook up some new AI video with Google's VO2. So Google DeepMind has announced the cost structure and availability for its VO2 video generation model. which is now accessible via its cloud API platform. So creating a video using V-O-2 is priced at 50 cents per second, translating to about $1,800 per hour on the vertex cloud, as noted by Google Deep Mind researcher John Barrett. For context, the Blockbuster Film Avengers Endgame cost approximately $32,000 per second to produce using traditional methods.

So yeah, $32,000 for a very highly visual blockbuster film like Avengers Endgame, where right now it's 50 cents per second for V-O-2. Obviously, those two things are not the same, right? Don't think they are. I'm not trying to draw a comparison, but you know, you might be thinking 50 cents a second. Is that expensive? You know, how much do? Well, there you go.

$32,000 per second for high-end cinema. Yeah, I called Avengers Endgame High End Semina. I like it. I watch Marvel movies. I have hobbies, right? It's almost like I'm arguing with myself as I don't sleep and just read and talk about AI all day. I'm like, yes, I'm a real boy. I have hobbies.

All right. So although VIA2 is more expensive than Open AI's SORA, which charges $200 per month with no usage cap, it remains a more affordable alternative to conventional film. making. So the pricing only covers the AI generation process, but additional costs, obviously, for human labor and multiple iterations to achieve desired results should be considered. Yeah, so obviously a human still needs to go in there and work with this. So that's not the total cost, right? This is just the cost to actually produce it. So AI enthusiasts and professionals could consider the potential of AI models like B02 to completely change video.

productions and its implications for cost savings and efficiency in creative industries. So aside from the API, V-O-2 has also just been released across a variety of platforms. So we talked about it last week. It was released in a limited capacity in YouTube shorts. You couldn't use the full V-O-2 model, but now you can. So you don't just have to cook it up with your own developers using the API because now free pick and F-A-L. I don't know if that's fall A-I or F-A-I, right?

But FreePick and F-A-L also have this service baked into their video generation platform. So this is pretty exciting news because, well, number one, I'm surprised, right? I'm surprised that Google did not make this a video on its front end or did not make this V-O-2 available on its front end for the general public yet. I'm assuming that will come at some point once they figure out pricing, but they did make it available via the API. And what that means is just about any video generation company out there that uses multiple APIs as an example, free pick and fall or FAL, right, they can start using V02. And why is this important?

Well, VO2 is hands down the best AI video model out there. It is better than Open AIs SORA. It is better than Kling out of China. It is better than Adobe Firefly, right? And I'll say it's not even necessarily close. So the VO2 model is extremely impressive. And I will say this, SORA, we got a tease of it first, right? And then we had to collectively wait like eight months.

But at least it's now available for anyone inside OpenAI's platform when you go to SORA.com. So it's interesting that Google didn't release this. on its own platform first, right, where you can like log in and use it, right? You have to use it via the API or one of these other third-party service providers, or you can use it in a very limited capacity in the YouTube shorts section. But it is by far the best AI video model, and I will say it is the first AI video model that I think will confuse the general public, right? Because I think with, even with Open AI's SORA, which I would say is probably in

second place. I think there's, you know, some competition there for who's next best after Google's V-O-2. But I'll say with SORA and a lot of these others buying for second place, for the most part, you can tell, right? If you just take your first shot out of SORA, you can usually tell it's AI generated, right? It struggles a little bit with physics. It struggles with understanding real-world simulations. Google not as bad, right? It's not perfect. Right. I think there was one kind of viral example that once all these platforms were released, because there are some trusted testers that did get early access to VO

and can go use it inside Google's platform. But there was one kind of famous or viral comparison video. It was someone cutting tomatoes, right? A close-up of someone cutting tomatoes. And SORA and all the other, you know, some of the AI video generators from China and, you know, runway and all these other.

others really struggled, right? And sometimes it was cutting a finger and sometimes the tomato just kind of cut itself or, you know, you'd keep chopping the tomato and it just wouldn't come out into little slices. And then the V-O-1 was extremely impressive, right? So I do think this is the first video that will confuse the average human viewer in terms of it's real, right? And yes, you can start with an image as well. So you can get a, you know, use some of the best platforms out there, like Mid Journey, get a, get it still image and then create a video that looks very realistic. So there's obviously some great upsides to this, right? Like maybe corporate training videos are going to get updated, the ones that haven't been updated since 1997. And they're

still like a three by three aspect ratio, right? And they're terrible. Right. So maybe there's, there's good use cases like old, you know, companies are going to be able to create more engaging content, they're going to be able to update more videos at a cheaper cost. And smaller companies are going to be able to produce high quality videos that just didn't have the budget or the talent or the expertise to create before. But obviously, there's a ton of downsides with this because I think people are not going to be able to realize what is real and what is fake, especially when the next version comes out. But I do think with enough generations and in the hands of a skilled person, it can be hard,

at least in short little bursts to tell if VO2 is real or not. All right. Our kind of our last piece of AI news. Well, Microsoft has a quantum leap. So Microsoft has announced a significant milestone in quantum computing with the unveiling of its myerana one chip, which could revolutionize industrial scale problem solving. So also they said they discovered a new state of,

matter, but we'll leave that discussion for like the scientists. This is an AI podcast, right? I don't, I don't understand that part. But the announcement led to a boost in shares for quantum computing companies with D-Wave quantum up nearly 10% in Rigget Teddy computing rising 2.5%. I got that wrong. Rigetti computing, right? So the whole quantum computing industry with this Microsoft announcement just went boom over the weekend. So Myerana 1 is a quantum computing unit or a QPU. Yes, I know. Now we have CPUs, GPUs, NPUs, TPUs, and QPUs.

I guess we're just going to PU everything until we run out of other letters in the alphabet. So Myerana 1 is a quantum processing unit or QPU that utilizes a new type of qubit to the top. apological qubit described as small, fast, and digitally controlled. So Microsoft claims its architecture can potentially fit a million cubits on a single chip surpassing IBM's goal of a 100,000 qubit quantum supercomputer by 2033. So yeah, Microsoft said IBM, hold my QPU, and they just said, we're going to 10-ex it. So the development marks a transition from scientific exploration to technological innovation after Microsoft has been researching this for apparently more than 17 years. So quantum computing is expected to transform, well, all fields.

And that's why it's important if you're following Generted AI. But it's especially expected to transform fields like pharmaceuticals, cybersecurity, and supply chain optimization, although mainstream applications are still. years away. But despite enthusiasm, industry leaders like NVIDIA's Jensen Wong and Mattis Mark Zuckerberg cautioned that practical quantum computing is still a long way off. So quantum competing is expected to revolutionize fields such as, like I said, pharmaceutical, cybersecurity, and AI, although it's going to be a while until we actually see this into fruition. So we don't cover quantum computing a ton on this show. We did cover it once.

But here's why it's important for AI. Think of it like this. It will make everything thousands times faster. So think of all the power in the compute that is needed right now to train new models and when we actually use them. If quantum computing comes to fruition, right? And if this new myerana one chip does help in that quest, everything will be like, I mean, according to reports, up to a million times faster. Because again, I'm not an expert in this, but how traditional computing works is you essentially have one part of a computer working on one task at a time, right? It can't work in parallel.

You know, one part of a computer works on one part of the task, then the next part of the computer works on another part of the task. Whereas this with this new myerana one, reportedly, it'll just have like a million cubits working on every single possible explanation or part of a problem in parallel. Right. So things that would literally, in theory, take thousands of years, could be done in seconds when and if quantum computing is. achieved. So this does not mean we've achieved quantum computing, but this is a pretty big milestone and a quantum leap on the quantum quest from Microsoft. So we probably won't be covering this too much as it's not like super generative AI, but it is something that impacts literally everything. Because then instead of having to use 100,000 GPUs or 200,000,

thousand GPUs over the course of a year, right? It's a very time-consuming, energy-consuming, costly process. In theory, you could do all of that training, async, right? You could run it simultaneously and probably get it done in, I don't know, like a minute or something like that, right? Again, is that science fiction? Maybe a little bit, but it's starting to look like more near-term science fact than science fiction. So pretty interesting. All right, so let's wrap it up and say there might be some new announcements, maybe as soon as this week.

So according to the Great Vine and the Great Vine is, you know, just people on Twitter and reporters on the Internet. Anthropic may be releasing a new version of Claude this week because tipsters online are saying that they've seen references of Claude 3.7 Sonnet, which would be Anthropics' newest model, which come on Anthropic, like ever. everyone was saying, you know, they updated Sonnet 3.5 and they just called it Sonnet 3.5 new. And everyone's like, yo, just call it Sonnet 3.6 or something, right? But apparently, we may be seeing an announcement on Claude 3.7 Sonnet, which would have reasoning capabilities and kind of that extended thinking for detailed step-by-step problem solving. and it would offer users, reportedly offer users the choice between quick responses and a thorough analysis, making it ideal for AI agents, complex workflows, and customer interactions.

So according to rumors, Claude 3.7 Sonnet is Anthropics' most intelligent model to date, and the first to offer extended thinking. So kind of this hybrid model situation, right, where it uses a little bit of the quote-unquote old-school transformer, but then it also uses a reasoning model. So it is reported that Amazon could be announcing this or Anthropic could be announcing it in step with Amazon Wednesday at the Amazon AI Alexa event. So we'll be covering that this week.

And also there's rumors that, you know, whether it's right around the same time or in the weeks after OpenAI might be releasing their new GPT 4.5 technology shortly. after Anthropic releases their new hybrid version of Claude. So again, this is all rumors and rumblings, but we are beginning to see some kind of snippets or breadcrumbs of this in code. Right. So I believe this was spotted on Amazon Bedrock system,

showing Claude 3.7 as a model choice and then was reportedly taken down. So we'll see it's rumors and rumblings for now, but AI does not sleep. And maybe I should start sleeping a little more. All right, I hope this was helpful, y'all. If it was, please repost this, right? So if you're listening on YouTube or Twitter, please don't just keep everyday AI as your secret cheat code. Right.

Our team spends so much time every single day, making sure you, dear listener, are the smartest person in AI at your company. So when someone's like at your company is like, hey, should we be using Brock? Right. We've been using Open AI for three years and they're like, let's get off Grock. You at least know, hey, there's a couple of things we should be considering in this conversation. Our goal is to make you the smartest person in AI at your company so you can grow your company and grow your career with generative AI. So if that is you and if this was helpful, please consider sharing this. If you're on the podcast, thank you for listening. Please subscribe to the show.

Leave us a rating. We'd appreciate that. Also, go listen to episodes 443 through 447, our 2025 AI predictions and roadmap series. Thank you for tuning in. Go sign up for the newsletter, Your EverydayAI.com. 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.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.