Everyday AI Podcast – An AI and ChatGPT Podcast - EP 586: OpenAI releases GPT-5 in ChatGPT, Google’s impressive Genie 3 and more AI News That Matters

Episode Date: August 11, 2025

OpenAI released GPT-5, and it's.... polarizing?Google dropped something kinda outta this world.And Anthropic picked a bad week to drop a new model.This week was one of the busiest in AI of the year. If you missed anything, this is your one-stop shot to get caught up. On Mondays, Everyday AI brings you the AI News That Matters. No fluff. No B.S. Just the meaningful AI news that impacts us all. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on this? Join the convo and connect with other AI leaders on LinkedIn.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:OpenAI Releases GPT-5—Smarter, Faster ModelGPT-5 Integration in Microsoft Copilot, AzureApple Intelligence Announces GPT-5 IntegrationGPT-5 Multimodal Input and Output FeaturesGPT-5 Rollout Issues and Model Router BugsAnthropic Launches Claude Opus 4.1 UpdateGoogle Genie 3 World Model DemonstrationOpenAI Debuts GPT OSS Open Source ModelGoogle Gemini Guided Learning LaunchesEleven Labs Releases AI Music GeneratorMeta Forms TBD Lab for Llama ModelsChatGPT Plus Plan Rate Limit ControversyUser Backlash Over Removal of Old ModelsCompetition Among AI Model Providers EscalatesTimestamps:00:00 GPT-5's Global Impact Unveiled03:22 "GPT-5: Stellar Yet Polarizing Release"06:23 "OpenAI's Impactful GPT-5 Update"11:51 "GPT-5 Integration Expands Microsoft Reach"13:19 Microsoft Integrates GPT-5 in AI Tools17:15 "GPT-5 Surpasses, OpenAI's Model Looms"23:18 "Guided Learning with Google Gemini"25:26 "AI Integration Critique in Education"30:40 AI Industry Disruption by GPT OSS34:49 AI Advances: Genie 3 Unveiled37:54 AI Video in World Simulators42:23 ChatGPT Plus Users Gain Higher Limits46:36 Altman on Unhealthy AI Dependencies49:41 Tech Updates: New Releases and Controversies51:24 Tech Giants Launch Major AI ModelsKeywords:GPT-5, OpenAI, AI news, large language model, ChatGPT, Microsoft Copilot, Apple Intelligence, iOS 26, multimodal model, model router, reasoning models, AI hallucinations, factual accuracy, AI safety, customization, API pricing, Anthropic, Claude Opus 4.1, agentSend 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. OpenAI has released GPT5, and it's not just going to impact their 700 million weekly active users,

but billions of other users across the globe as it starts to roll out to Microsoft co-pilot and even Apple's users as well. That is the headlining. piece of news in an extremely busy week in the AI world. As someone that does this every single day and recaps the AI news that matters every single Monday, I'd say this has probably been the most consequential week in AI news since December of 2024. So we have a ton to cover, not just GPT5 and those implications all across the board, but also

Google's new impressive. Genie 3 and Anthropic picked a pretty bad week to release a new model. All right, we're to be covering those stories and a whole lot more today on Everyday AI. What's going on, y'all? My name's Jordan Wilson. Welcome to Everyday AI. This is your daily live stream podcast and free daily newsletter helping everyday

business leaders like you and me, not just keep up, but how we can get ahead and use all this information, leverage it to grow our companies and our careers. So you could waste hours every single day trying to keep up. Or you can read our free daily newsletter. It takes about six or seven minutes every day. Or join us on Mondays as we bring you the AI news that matters. All right, there is a ton to get to this week. Let's dive right into it.

Hey, live stream audience. Thank you for joining. If you're on the podcast, FYI, we always keep a, put a link to today's episode in our newsletter. We're sharing a couple things on screen, but not that we can't hopefully accurately convey in the in the podcast so thanks for joining us michel on the youtube machine and bronson big bogey face as well thanks for joining us geordy and charles on the lincoln machine good to see everyone happy monday brea joining us from chicago holding it down for my hometown uh gerald joining us from uh san diego robert tuning in from porto de rico santiago chili in the house hose

Let's get to it. So first and foremost, the biggest AI news, not just of the week, but probably of the year, just because of its huge implications across the workforce. So Open AI has released GPT5 as its most capable in fastest model yet. So Open AI has released GPT5, a major upgrade that delivers smarter, faster, and more useful, performance across many tasks. And it's been not all good. There's been quite a few bumps on the road so far, even though it's only been out for about four or five days. So we're going to have a separate news piece later talking about the bumpy start. But so far and on paper, GPT5 is stellar. And for whatever reasons, it's a polarizing model. And I think I have some thoughts on that for later. But let's

it to the good stuff. So Open AI says GPT5 offers significantly improved reasoning and built in thinking, making it better at complex, multi-step problems that previously required human expertise. So Open AI says that GPT5 is faster and more efficient, which should reduce latency for developers and product teams integrating advanced AI apps into their workflows, as well as for everyday users using this new model on the front end. chat gpti.com and it has rolled out to both free and paid users. So yeah, now if you go to chat gpt com and log into your account, even if you have a free account, you will have access to the most powerful and newest model that is, it's technically a hybrid model because it uses this

auto routing functionality that decides if your query should just be kind of resolved or answered immediately or if the model should think about it a little bit. So even for free users, now you have a capable model that can switch. So there were quite a few problems with Open AI's rollout, but one of them was reportedly this model router that is kind of the magic in the secret sauce, but behind this new GPT5 architecture was kind of broken or not functioning correctly in the first day or so after release, which led to a lot of kind of negative press. But let's get to a little bit more what's new in GPT-5.

So it supports broad multimodal inputs and outputs, enabling richer interactions with text images and possibly other media in the future. So right now, no video inputs, which we had thought that it was going to have, at least when we started hearing the GPT5 rumors long ago. But no video input or video outputs, unlike. what we have in Google Gemini 2.5 Pro. So OpenAI highlighted the improvements in factual accuracy and honesty, aiming to reduce confident but incorrect answers or hallucinations,

as we all call them, and to make responses more reliable for real-world decision-making. Also, OpenAI emphasized safety and guardrails, and they built comprehensive mitigations to limit misuse, reduce sycifancy, and refine conversational tone and behavior. So the new GPT5 update introduces more customization as well. Letting users you can kind of choose different to pre-built personalities. There's some small little UIUX things like being able to customize the color of your chats.

So I think in some regards, OpenAI has gone a little Apple with this update because, you know, kind of, I don't know. For me personally, I don't think we needed these kind of four personalities that you can chat with or color customization, you know, essentially changing the color of your conversation. So it's not just a bunch of white text blocks. But regardless, for developers, it's huge because GPD-5's API pricing, if you are building on top or if your company is wrapping around OpenAIs technology, It is ridiculously cheap, right? And I think this is really going to put the squeeze on Anthropic and others. So Open AI notes that applications across code, creative writing, health and enterprise productivity have improved, which could reshape hiring needs as well by shifting routine expert work toward AI-assisted workflows.

So yeah, we're going to have a lot more on this story today. But also, if you do want all of the deeps, details. Make sure to check out Friday's episode. That's episode 585 when we go over all the benchmarks, all the specs and also seven big trends that I think people need to know. So make sure if you haven't already, go listen to that episode 585 for more info. What do you all think so far on the GPT live stream audience? I'm curious. It's been very polarizing more on that here. I think our last story for the AI News That Matters Roundup today is a little bit more about that. But I'd love to hear your thoughts.

I was kind of surprised. So CEO Sam Altman put out a bunch of tweets over the weekend. So one of them was talking about just how few people previously before GPT5 were using the reasoning models, even plus subscribers. I believe it was only seven. I'd have to double check on that. try to do that live. Yeah, this podcast is unedited, unscripted. So sometimes we take short turns here. So yeah, so Sam Aldman did put out a tweet over the weekend. He said the percentage of users using reasoning models each day is significantly increasing.

He said, for example, for free users, we went from previously less than 1% using reasoning models, even though free users did get a few reasoning queries each day. to 7%. And for plus users, so those are users paying $20. It went from previously only 7% of users using reasoning models to now 24 with this kind of smart auto router. So to me, it's absolutely wild that people that were paying $20 a month, only 7% were using reasoning models, right? I mentioned this last week on the show. I personally don't use GPT4.

or didn't use GPT4O, I would always use a reasoning model as soon as the day they came out because they're just much better. I think people are impatient. They don't want to wait. Yeah, sometimes using a reasoning model, including GPT5, especially if you toggle the setting on to use more thinking, essentially. It might take a couple of minutes. But I was kind of honestly flabbergasted by that little stat that said,

And yeah, only 7% of users are using it. But right now, it's Monday morning. Chat GPT is slammed. It's currently down. So there's some breaking news. It wasn't down 10 minutes ago. All right. Our next piece of AI news.

Yeah, a lot of GPT5. So Microsoft has also rolled out GPT5 into co-pilot, GitHub, and Azure. So Microsoft, literally, as soon as GPT5 was a number, announced did also announce that they rolled out GPT5 into copilot, GitHub, copilot visual studio code and Azure AI Foundry. And I believe it's in their AI agent builder as well inside co-pilot studio. So the single most newsworthy change is that Microsoft is using that real time router to choose between the best GPT5 model for each task automatically.

So essentially, you know, if you are on a paid plan, you have these different tiers of thinking that you can use. So there's no longer, you know, alphabet model soup inside of chat GPT where you could choose, you know, previously, you could choose GPT 40, GPT 41, GPT 41, GPT 45, 03, 04 mini, 04 mini, high, 03 Pro, right? So you no longer have these multiple tiers of non-reasoning, in reasoning models. You just have GPT5 with different levels of think,

which is essentially spending more time, more compute to deliver a better answer. So Microsoft 365 copilot users, as long as their IT department enables this, are going to see a huge, probably increase in their outputs as well, which is why I think this GPT5 is so much bigger than just those 700 million weekly active.

users, which that number is insanely large, right? ChatGPT has the most active users for any AI chatbot by far. But when it comes to the enterprise, especially here in the U.S., but I'll say the world, the world runs on Windows, right? The world runs on Microsoft. So the fact that literally you have millions, more than 100 million paid co-pilot users in the world, I think was the last number I saw on August 1st. So if you are a paid co-pilot organization, which so many are, you should have access to GPT5 automatic. So also on the front end, so even if you're not a Microsoft 365 organization, you can go

on their copilot. com website, essentially accessing Microsoft copilot online. And there's also a new free smart mode powered by GPT So in instances like literally right now as I'm recording, because I just went to chat gpt.com, and it's down, you can go to copilot.microsoft.com. And even if you're not a $20 a month Microsoft copilot user, which is different than the Microsoft 365 enterprise licenses, you can use that free GPT5 smart mode. So also, like I said, GitHub co-pilot and Microsoft's Visual Copilot Studio in paid plans, We'll get GPT5 for writing, testing, and deploying code. And Microsoft says the new models from OpenAI Excel on longer, more complex coding,

and end-to-end agenic tasks, and they can be selected as a model picker. In the Azure AI Foundry, it makes all GPT5 models available for developers. That's pretty big and uses a model router to pick the optimal model per prompt based on complexity, performance needs, and cost efficiency. Enabling mixed model orchestration in production workloads. So yeah, what does this mean? Essentially, whether you're using chatGPt.com, whether your organization is a paid Microsoft co-pilot organization or like thousands of companies, thousands of

SaaS products use GPT5 as well, right? I obviously use a ton of AI tools. And by the time I didn't even notice on Friday, many of them had already ported over to GPC5. So what this means aside from if you're a Microsoft heavy organization is probably just hundreds or thousands of the tools that you use are now going to get smarter. Speaking of getting smarter, could Apple finally get smarter? Well, some reports are saying maybe so because some new reports. says that OpenAI's new GPT5 will become available inside Apple Intelligence when iOS 26 is released in early September. So yeah, maybe Apple will finally, for its billions of devices across the globe,

maybe there will actually be something smart or something resembling AI inside of their phones. So Apple reportedly said that the GPT5 integration will roll out. with iOS 26, iPad OS 26 and MacOS Tahoe in early September. You do have to have a newer device for this to run because it does run on device. Well, some things are still going to be sent to chat GPT more on that here in a second. But the biggest immediate change for users is that Apple intelligence will hand off difficult queries to GPT5 when Apple's own models that run on device cannot. answer them. So more Siri and system prompts may invoke GPT5 behind the scenes. And on Apple devices,

GPT5 will be used in features beyond just Siri, including Apple's writing tools and their visual intelligence, which is their camera-based assistance. So that's expected to tap GPT5 for richer responses and image understanding. So this partnership reportedly will give Apple a quicker way to improve on device intelligence while the company reportedly develops its own AI chatbot. We talked about that last week, but I wouldn't have any faith in that coming anytime in the next couple of years. So pretty important, right? And that's why we say when we're talking about GPT5, even though I think it's gotten a mixed

reaction so far, because it depends on if you're using the thinking or not. I think if the model auto routes or if you're not overriding and enabling the thinking by default, you might not be super impressed or super happy with GPT5's outputs, especially, well, number one, when that model router was broken once it launched, but also if you've been using Gemini 2.5 Pro, which has until Gemini or until GPD5 overtook it on standard third party benchmarks on LM Arena. So now GPD5 is the highest rated model by users and by many third party benchmarks, including artificial analysis, you know, aside from just the scientific ones, I still think in some instances, you're not going to know the difference, right?

And I said this, I said this last week. I said that actually Open AI's open model, which we're going to get to here in a minute, is going to be bigger news ultimately than GPT5. And, you know, as it turns out, I do think that will hold true, even though GPT5 is going to bring a higher level of Intel. to billions of users across the globe just because of, like I said, their partnerships and integration with Microsoft and with Apple. Bad timing for Anthropics.

So our next piece of AI news, Anthropic released an update to Claude, rolling out Claude Opus 4.1, what they're calling an incremental update focused on improving agenic tasks, real world coding and reasoning. So Anthropic released Opus 4.1 just days before GVT4.5. So talk about bad timing. Or maybe they wanted to slip this one in under the radar with not a lot of fanfare

because it was just an incremental update. So on some important benchmarks, I think Anthropics models have obviously found their home with software developers, coders, et cetera, people on the dev side. So Opus 4.1 achieves a 74.5 in software engineering accuracy up from 72.5 in Clod Opus 4 and 62.3 in Quad Sonnet 3.7. So again, just incremental gains. And maybe that's why Anthropa kind of slipped this one in in a busy news week. But the company did highlight improvements in in-depth research, data analysis, detailed, track. and agentic search areas that help the new model manage multi-step workflows and code focus tasks.

What we didn't see is a price improvement. It is still just crazily expensive, especially compared to OpenAIs GPD5 and Google's Gemini 2.5 Pro. So Claude Opus 4.1 is available. If you log on to claw.a.i, you can. can use it inside Claude, inside Anthropics API, and through other partners such as Amazon Bedrock and Google Google Cloud Vertex AI. So I don't know. What do you guys think? I tried it out, obviously, when it was released. There's some instances, yeah, if you need to code something,

build a dashboard, but I'll say for the everyday, you know, the everyday business leader, I don't think you're going to be using Claude 4.1 Opus a ton, especially infropic is infamous for their terrible rate limits, even if you are on a paid plan. So their base $20 a month paid plan, you know, you might get, I don't know, especially if you're working with longer context windows. If you're putting in a ton of info, there's been times where I have prompted Claude about five times and I hit my rate limit. So on Opus 4.1, I haven't tested it, but assume it's going to be pretty bad if you are really pushing it. So again, I wouldn't get too excited about this unless you are in software development, unless you are coder development, et cetera, software engineering. And you have maybe their max plan that $100 or $200 month.

But they also cut the limits back on that a couple of weeks ago as well. So I'm not sure if anyone's going to really be. going wild over Claude Opus 4.1, seems like bad timing, or maybe that was intentional. 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. All right. Our next piece of AI news, Google launched their new guided learning inside Gemini.

So this is a new Gemini experience. experiences, a new Gemini experience that uses the LearnLM model trained and multimodal content to help users build understanding step by step rather than just delivering quick answers. So if that sounds kind of familiar, if you're like, wait, I've heard of this. Uh, yeah. So Chad Chb-T last week released their new study mode. And this is kind of Google's version of that. So the core promise here is to promote active engagements by asking probing open-ended questions rather than just giving answers.

So guided learning inside Google Gemini delivers rich multimodal responses, including images, diagrams, videos, and interactive quizzes designed to teach the process, not just give the final solution. So yeah, obviously a lot of these updates coming before the fall rush of students here in the US going back to college. So obviously with chat GPT study mode, with GPT5 being free and pretty good, pretty good limits on the plus plan, which are improving more on that here in a second, their study mode. And now Google Gemini on their learning, their learning offering here inside guided learning, pretty impressive. All right. So at least in my little bit of testing

so far, I seem to like the interactivity and the multi-modality of Google's version a little bit more. So, hey, we did do the chat GPT version of this last week looking at their new study mode. So if you think we should do the guided learning, let me know, just maybe leave a comment, just say guided. I always just like to know what you all want to hear. Maybe we'll put a poll out in the newsletter if you're listening on Spotify. you know, you can leave a comment on the Spotify episode. I can't reply to them, but just say guided if you care.

If not, that's fine. So a little bit more. So Google reports that the capability was developed from years of research and partnerships with external experts and students. And that simple improvements to prompting were insufficient, which prompted the design of LearnLM. So a little bit more for educators, Google created a shift.

yearable link for direct use in Google classroom, positioning guided learning as a classroom partner that supports active construction, constructive learning rather than replacing instruction. All right. So I love what Google's been doing and Open AI, just getting more AI into the classroom because I've talked about this way too much. And I'm not going to go on a little rant here. But students who are graduating in the U.S. over the past two years are illy unprepared to go out and make a difference in the work world

because the majority of universities since 2022 have been completely shutting down AI use, not teaching it, not promoting AI literacy. And obviously every single employer for the fewer jobs they are hiring for now, especially on the entry level side, they want people who are AI literate. And that is for the most part, not students because all students have been doing inside of chat GPT or Google Gemini is having it write their papers and they copy and paste it and they're not actually learning for the most part how to use this technology.

So I like the move from both OpenAI with their study mode and guided learning in Gemini to hopefully help make learning a little stickier and to also at the same time show off some capabilities of large language models that copying and pasting your history final 10 minutes before it's due is not going to teach you. All right. Our next piece of AI news. And like I said last week, this could ultimately have the biggest long-term ramifications. So Open AI has released GPTOSS, a new free and powerful OpenWeets model.

So there's two different versions, a 20 billion parameter and 120 billion parameter models. These are open source models. You can download so you can run them locally on your machine, cut off the internet. You're not sending any data back to OpenAI. You can fine tune these or your company can fine tune these as well as they are Apache 2.0 license models. So that means you can commercialize them. You can create literally companies based on this extremely powerful models.

And for the most, I'll say they kind of fell somewhere around a GPT4.0. level or maybe an 04 mini level, right? So they're obviously not as powerful as GPT5 or as open AI's previous best model. They're 03 model, but they're fairly powerful, especially for a free open weights, open source model that you can download fork and build a company on top of. So according to open AI, the smaller 20 billion parameter model

scores an 85.3 on the MMLU, which is kind of an old school ACT type test for large language models, putting it in the ballpark of recent GPT4 level models and demonstrating some real reasoning capabilities. So the Apache 2.0 license allows for unlimited commercial use, fine-tuning, redistribution, and patent protections, meaning that companies can embed, modify, and sell products using the new GPT OSS models without fees or geographic limits. So if you want to run the bigger version, you're probably going to need a supercomputer or like an H-100 chip from Nvidia. But for the most part, if you have a newer laptop, you should be able to run the 20 billion parameter. If you have 16 gigabytes of RAM and a powerful enough GPU, you should be able to literally

go download this, run it offline, fine tune it. And this really changes, I think, more than anything else. So not just individually, not just individually for companies that maybe don't want to keep paying API pricing, right? So you're not going to get state of the art results from GBTOSS. But let's just say that you've been using a model like Claude 3.5 Sonnet, which a lot of companies are still using. or maybe you were previously using GPT40, right? There's probably some instances that, hey, you might not want to use GPTOSS for everything, but for maybe some of the more basic, less complex or highly sensitive cases that you might

want to use an AI model for. This is a perfect use case. I do know that many companies are going to save millions of dollars a year that they were previously spending or by getting off and more expensive. expensive API provider like Anthropic, right? So yeah, maybe what I know a lot of companies are doing when they roll out API access or they're using an API within their organization, they're just choosing one, right? So yes, it might be worth paying that extra premium price to use Claude's API for some of your

people, you know, your software developers, although I do think GPT5 is going to be better. There's a reason that cursor made it the default model inside of it. It's, you know, vibe coding platform. But so many companies are just picking one provider and then rolling that out for their entire organization. So, you know, in this case, GPTOSS is more than capable for some of those more basic things like customer service, writing, using some basic vision tasks. So I do think in the long run, GPTOSS is going to completely shake up the industry because, I call them mid-tier. This is going to push mid-tier providers

or those people that aren't like 1A, 1B, like Open AI and Google. It's going to push everyone else to bring their prices down and to have more capable models. So this is an extremely seismic shift in the future of AI development. I said when I cover this, and if you do want to go listen to this,

I suggest you do go listen to episode 584, where we break this down in much more detail. Essentially, opening I went scorched earth. So yes, they kind of, I wouldn't say they fumbled the GPT5 rollout, but they had some mistakes in the rollout. Talk about like, you know, they had some basic charts that were all wrong, but I'm like, yo, humans hallucinate. Anyways, between GPT5, making GPT5 free with some improved limits, more on that here in a

minute. And then in open source GPTOS, this is monumental in terms of. what it means for companies that either couldn't afford to have state of the art AI or for those companies that were probably overpaying and looking for a new partner. So now I do think this is going to squeeze a lot of those mid tier. Either it's going to make it better for all of us or a lot of these companies maybe are just going to go out of business or they're going to have to actually start charging more for some of their base use cases just to try to make ends meet.

All right. Our next piece of AI news, more acronyms from Meta. Can't wait. So according to the Wall Street Journal, Meta has formed a new team called TBD Lab to lead development of the next version of its Lama large language model. So TBD Labs reportedly sits within MSL, which is Meta's superintelligence Labs group. All right. So MSL is kind of the umbrella lab and the new TBD lab will be working on the next

meta-Lama models. And it was created to coordinate foundation models, fair, research, product work, and next-generation model development. So the team reportedly includes at least 18 researchers hired from OpenAI, additional hires from Google, and nine meta-employees moved over from internal infrastructure. So former scale CEO, Alexander Wang, will oversee TBD Lab after joining META via a $14 billion stake in his former company. In a memo, he said parallel collaborations are already speeding progress. So META has been on an aggressive hiring fee since early June,

with reports that Mark Zuckerberg personally recruited candidates at his homes in Lake Tahoe and Palo Alto. So industry-wide competition for AI talent has driven unprecedented competition with some packages reported already exceeding those of NBA stars. So yeah, we saw people are getting multiple hundred million dollar comp packages. At least one was reportedly worth more than a billion dollars over four years. So this is pretty big. So when you see TBD, it's not to be determined. apparently this is the new lab that sits under meta's kind of super intelligence umbrella so yeah apparently it just looks like they're not going to have their quote unquote

super intelligence team building the meta models it's going to be this TBD lab specifically that will be working on what we would assume is going to be meta llama 4.5 so next piece of AI news which visually is the most impressive thing I've seen in a very long time. Last year Monday, did anyone check this out when we shared it in the newsletter last week? Google's Genie 3 is bonkers. All right. So Google DeepMind released their Genie 3 world model, a world model that generates live, open-ended virtual environments signaling rapid progress toward AI systems that can predict

and react like physical reality. So Google says Genie 3 can generate environments on the fly, completely on the fly, from a single text prompt and update them in real time as users interact, a stepped beyond pre-built game worlds. The model also introduces what they're calling world memory, preserving persistent changes over time. For example, the one that Google Gemini showcased. a painted wall in one room.

There's a couple brush strokes. In the demo, they left the room. They came back and the brush strokes were still there. So world models, if you're like, what the heck are these? Why do they matter? Well, they're designed to reason about physical mechanics. So these are going to help embodied AI, right? When we talk about humanoids, AI robotics, you know, large language models,

essentially they have all their unstructured data and structured data from the internet and from humans right uh it's much slower and a much more time consuming and costly process for embodied a i or a i that ultimately lives outside in the real world to understand how the real world works so that's why something like a world model like genie three that has understanding of things like gravity, light, shadows, how two objects when they run into each other, how they interact. It's actually huge, right? Obviously, there's great applications, right? Like hopefully getting safer autonomous vehicles on the road, video game development,

movie development, being able to, you know, scout and understand new locations for, I don't know, big architecture projects, logistics, etc. So I think there's many far-reaching use cases of a model like Jeannie 3. So OpenAI's SORA previously, so that's their AI video model from OpenAI showcase physics-aware video generation. And Google's V-O-3 can now generate an eight-second video from a single image, highlighting intensifying competition in AI video and simulation. So that's why there is a strong connection between AI video in obviously these world simulators. And actually the very cool kind of example that we shared in our newsletter last week was someone created a video inside Google Vio, kind of like an overhead drone flight. And then someone that had access to this new Genie 3, which is only tested trusted tester users right now.

So it's not publicly available. They actually took that AI video, put it. it into Jeannie 3.0 and took it over from there and then kind of move the camera control. So you can kind of pause at any point, point in the camera somewhere else and then go explore. And it generates it in real time. Right. So obviously a lot of people, when they think of these world models, they just think, oh, it's for video games.

It's for movies. It's for, uh, you know, AI video. And yes, but it also helps embodied AI, AI get much more data on how the real world operates because as an example in Google, Genie 3, right? If a care, if someone is walking along a path, you know, there's ripple effects in the water, right? That helps train future models to become better. So potential use cases include training for dangerous scenarios such as disaster response where simulated environments could help first responders build skills in muscle memory without real world risk. Also, education

sectors could benefit through immersive, interactive, visual learning experiences that adapt to student actions, potentially improving retention and engagements. So the debate, though, continues whether these systems actually understand the world. But Google DeepMine defines world models as systems that use their understanding to simulate and predict environments while experts disagree on whether this counts as true understanding. Regardless, it's really freaking cool. We'll link it again in the newsletter, so make sure you go check it out. We had new updates across all sectors.

So we had, you know, GPT5 and Claude. On the world model side, we had Jeannie 3, which is pretty huge. And 11 Labs, traditionally known as a text to speech platform, just launched in AI music generator. So 11 Labs announced a new AI model that generates music, which they say is cleared for commercial use. Pretty big step there. So the company did share examples, including a synthetic rap track with references like Compton to Cosmos, underscoring cultural concerns about AI mimicking artists who live the experiences that the new AI model imitates.

So 11 Labs is moving into a legally complex space as the two big players in the AI music generating scene, which are Suno and UDio. were sued by the recording industry last year over alleged training on copyrighted music. It are now reported pursuing licensing deals. Reportedly, Suno and UDO are pursuing licensing deal with major labels. So to address this, 11 labs announced licensing agreements with Merlin Network and Cobalt music groups, which together represent some major artists and catalogs. So Cobalt says artists must opt in for AI licensing.

So the company's broader push follows its growth from a text-to-speech platform into a conversational bot and multilingual speech translation service, positioning 11 labs as more of a full-stack audio AI player versus what they've traditionally been known for or known as, which is kind of just a text-to-speech leader. All right. our last piece of AI news. Yeah, there was some backlash after chat GBT announced GPT5. And there's already been some swift actions and updates to the different plans since that backlash.

So Open AI is already saying that they are going to increase chat GPT plus rate limits for the new GPD5. thinking model following user criticism. So Sam Altman said on Twitter that rate limits for reasoning are being significantly raised for plus users. Those are people paying $20 a month after essentially, you know, when you broke it down, plus users were actually getting way fewer, way fewer messages with reasoning, even though only 7% of users were using it. So essentially when you had six or seven models to choose from,

and each of those models had their own rate limit, you just had way more usage ability. So when they kind of consolidated these models and got rid of the old ones, more on that here in a second, essentially usage went down by a lot. So St. Maltman did say that they're trialing, so this is not confirmed, but they're trialing to see if they can make this work,

a 3,000 per week message cap on GPT5 thinking for plus users, which I think would go a long way. And I think very few plus users would go past that. And if so, it's probably time to upgrade to that $200 pro plan, which has unlimited GPT5 usage. So the upcoming change follows widespread frustration after GPT5 launched with only 200 messages per day. limits for chat GPT plus users, which many said felt like undercutted the value. Also, OpenAI is adding a new user interface indicator to show which model answers a prompt because you don't really know, right? There's kind of this model router working under the hood.

So you don't know if you're using the quote unquote normal GPT5 or if you're using the thinking variation. So that should be cleared up soon. the company said. But I would say the biggest one, it wasn't just the limits in the messages that were the biggest uproar, as some of the loudest complaints online

were due to OpenAI pulling access to older models without warning for most users, including the very popular GPT40 model. So OpenAI initially removed and essentially retired these older models to make way for the new kind of model architecture with GPT5 being the main model. But this triggered tons of user complaints online, essentially getting rid of models and not really having a choice.

So users criticize GPT5's more direct tone and writing compared with older models that they no longer had access to like GPT41, GPT45 and 40. And they called GPT5 shorter and colder. So essentially, here's what was happening. So many people became reliant on GPT4 for certain things, right? Whether it was actual writing, doing things in their day-to-day job, or companionship, using it as a sounding board as therapy, right? A lot of people reported using Chad GPT as almost like a therapist when maybe their own therapist wasn't working well or they couldn't afford one. So a lot of people became kind of overreliant on GPT4. And Sam Altman actually addressed this on Twitter.

And he did say that they're considering keeping GPT40 and older models available for plus subscribers. So if you were a pro subscriber, you did have access to those old models right when GVT5 rolled out. But if you were a $20 plus or a free user, you did not have access to these older models. So I get where Open AI and Sam Olman's coming from, right? Everyone's complaining there's too many models, too many models, and everyone's been screaming, released GPT5. So, okay, you release GPT5, and that is the main only model. But then everyone still wants all the older models anyways, right?

Because they almost developed an attachment to them in an unhealthy way. So Sam Altman did on Twitter talk about over the weekend about how some users are forming kind of unhealthy relationships with AI models in general, and that Open AI will work to protect both user freedom, but also push back in edge cases to avoid harm dependency on AI or in instances where it's actually causing long-term harm for users who may struggle to distinguish fiction from reality. So essentially, GPT4 was number one, it was overly verbose

and they had to, you know, roll it back a couple of times in one case or in many cases, right? It essentially just mirrored or became a yes man or yes women for users. So no matter what kind of idea you threw at it or circumstances, a lot of times it would just side with the user. Like, oh, yeah, you were absolutely right, you know, and how you justified that decision that worker in your personal life where even if that was maybe bad. So I think essentially we have a problem. right? We have a problem and it actually took GPT5, which is better, right?

It's just better at, you know, strategy, uh, advice, writing in general. So I think a lot of people were becoming over reliant on the kind of, um, you know, this, this yes manning, uh, of these models that would essentially become a kind of dangerous echo chamber at times if you were using it for those purposes, uh, right? So we'll see what happens, but it does look like, number one, there's going to be more messages across the board. Number two, it looks like Open AI has fixed the model router, but it actually caused a lot of damage because by them not having that working, a lot of people's first impressions was, well, this isn't that great. It's not working that well, right? So because the model router functionality wasn't working 100% correctly when it was rolled out, a lot of the initial press and

reaction from GPD 5 was, oh, it kind of stinks. And the fact that people became overreliant on GPD 4, just kind of always mirroring whatever their assumption or opinion was in cases, maybe using it as a therapist, thought partner, or for companionship, whereas GPD 5 isn't really like that. It's a little more straightforward to the point, which I personally enjoy a lot more. So yeah, a big boge here on YouTube saying, if you want a therapist, co-pilot has a huge emotional intelligence. All right. Now, quickly, what's new and what's next?

So we couldn't fit everything into this update. It's already been 48 minutes. So let's wrap it here. So some bullet points of what's new and what's next. Google, big one. And this is like an hour old, just got rolled out. Google finally released its ultra-plice. plan for business workspace users.

It starts at $250 per month per user, but for the first three months, there is a promo that it's $106 a month. So this was previously available to Gmail subscribers, right? I'm an ultra subscriber in my personal Gmail, but didn't have access to it in my workspace or my work account. So this is pretty big. We'll probably be talking about this and maybe having a dedicated episode soon. Google's AI coding agent Jules has accident beta.

X-A-I's GROC-Imagine image to video generator was rolled out, and it's already in hot water for some unauthorized explicit images. I think it's terrible. Co-Pilot 3D is rolling out in Microsoft copilot labs. Cursor, released cursor, CLA, their command line editor to bring cursor to terminals. OpenAI's Codex platform got GPT5 access for paid accounts. Runway has released a faster version of its Gen 4 image model called the turbo. It's also cheaper as well.

Perplexity's Comets browser is demoing a max assistant option for more complex workflows inside of its agentic browsers. And reports are saying that Anthropics popular Claude code may be coming to the browser. All right, that's a ton. Let's very quickly recap the big news stories of the week. So first and foremost, Open AI has launched GPT5 as its most capable, fastest model yet. Microsoft is rolling out GPD5 across its ecosystem, as well as reports are saying that Apple will be rolling out GPD5 in Apple Intelligence in iOS 26 in September. Next, Anthropic launched Claude Opus, 4.1, with marginally better coding and reasoning, bad timing.

Google launched guided learning in Gemini to boost better learning just in time for back to school. OpenAI released GPTOSS, a free, powerful open weights model that you can download and use without the internet, without sending any data back to Open AI. Metas, according to the Wall Street Journal, has launched their TBD lab, which sits inside of their MSL and will be working, on meta llama models in the futures. Google DeepMind released Genie 3 world model, extremely impressive. 11 Labs launched commercial use AI music model.

And Open AI has boosted GPT5 limits after some pretty intense backlash. All right. That was a lot, y'all. There's going to be a lot more. If you haven't already, please make sure to go to your everyday AI.com. Sign up for the free daily newsletter. If this was helpful, please let someone know about it. Share this, post this.

If you're listening live on the live stream, please let someone know. If you're listening on the podcast, appreciate your support. Please make sure it takes 10 seconds. Go click that, subscribe or follow button on Apple Podcasts and Spotify. If you could, take 30 seconds, leave us a rating. That would mean so much to me. I'd appreciate that. Thanks for tuning in.

Please join us tomorrow and everyday 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, Premier 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.