Everyday AI Podcast – An AI and ChatGPT Podcast - EP 601: Nano Banana goes viral, why Meta’s AI could be in big trouble & more AI News That Matters

Episode Date: September 2, 2025

For the second time in just as many weeks, Google dropped impressive AI updates that their competitors will struggle to keep up with. Salesforce is cutting it support staff due to AI and OpenAI's backend tech could be one of its biggest updates yet. And that's just the start. Don't waste hours a day trying to keep up with AI developments. That's what we do.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:Meta Integrating Google Gemini & OpenAIMeta AI Losing Key Talent & ResearchersMeta Tightens AI Safeguards for TeensSalesforce Cuts 4,000 Jobs With AI AgentsOpenAI Launches GPT Real-Time APIMeta’s $14.3B Scale AI Deal in TroublePerplexity AI Announces Comet Plus for PublishersGoogle Adds AI Avatars to Google VidsStanford Study: AI Hurting Entry-Level JobsGoogle Gemini 2.5 Flash Image (Nano Banana) Viral ReleaseTimestamps:00:00 Meta Leverages Competitors' AI Models06:30 Google Challenges Duolingo with AI Update08:02 Google's Language AI Potential Unveiled13:48 Salesforce CEO on AI-Driven Workforce Reduction17:23 OpenAI's New Real-Time Voice API20:32 AI Technology Disrupts Customer Service23:55 Meta's AI Acquisition Challenges28:52 Publishers Question AI Revenue Plans31:35 AI's Dependence on Human-Created Data36:31 AI's Impact on Jobs and Startups40:19 Gemini 2.5: Multimodal AI Update41:59 Google's New Image Model Unveiled44:33 Tech Update: Flashcards, Rumors, and AI47:36 AI News and Insights ScheduleKeywords:Meta AI, Meta Superintelligence Lab, Meta layoffs, Meta internal policy, Meta data labeling, Meta AI chatbot, Meta Scale AI investment, Meta AI partnerships, Meta generative AI, Meta AI safety, Meta AI user protections, Google AI, Google Gemini, Gemini 2.5, Gemini 2.5 Flash Image, Nano Banana, Google Translate, AI language learning, AI practice sessions, AI-powered translation, multimodal models, AI image editing, AI character consistency, Google Vids, AI video editing, AI avatars, AI workspace tools, Salesforce, Salesforce AI agents, AI customer service, job autSend 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. Google keeps winning each week in AI.

Yeah, they had their mega viral nano banana release, but they had two other AI updates that slipped under the radar that I think are going to be just as impactful. Meta, on the other hand, is not winning. They're reportedly losing, not just in the battle to create super intelligence, but just losing their people. And I think Salesforce, I don't know if we can believe what they say anymore anyways.

There's been a lot happening in the world of AI. If you missed anything, don't worry. We're going to not just get you caught up. We're going to help you get ahead. All right. If you're new here, welcome. This is Everyday AI. What's going on, y'all?

My name's Jordan Wilson, and I'm the host. And this thing, it's for you. It's your daily live stream podcast. and free daily newsletter, helping everyday business leaders like you and me, not just keep up with what's happening in the world of AI, but how we can use this information, decipher it, know what's good, what's not to grow our companies and our career. So it starts here with the unedited, unscripted, live stream podcast.

But if you want to really take it to the next level, the cheat code is our website, your everyday AI.com. So there, you can not only sign up for our free daily newsletter where we're going to be recapping the highlights from today's show, but we're also going to give you. you every other thing you need to keep up because it's a lot. Speaking of keeping up, this is our normal AI News That Matters weekly segment.

We normally do this on Mondays bringing you the AI news that matters. But because of the holiday here in the U.S., we're doing it here live on Tuesday. So don't spend hours every single week trying to keep up. We're just going to tell you straight. Here's what matters. Here's what doesn't in the world of AI news because as any week, There was a ton going on. So yeah, if you're on the podcast, make sure to check out your show notes.

There's going to be a link to our website. So if you want to go check out more on these stories, we're going to be recapping them in the newsletter as well in live stream audience. Good to see you. Good morning to Dennis from New York City, Jose from Santiago, Angie checking in, Marie from New Jersey, Brian from Minnesota. It's great to see everyone. Let's get straight into it. So meta is using its competitors? Yeah, that's an interesting headline that popped up this week in AI news.

So according to the information, meta is exploring, integrating Google's Gemini and maybe even open AI's models into its meta AI assistant to deliver better text-based answers to user search questions. Yeah, so even though meta is spending reportedly billions of dollars on AI talent, the latest report from the information is saying they might just have to use their competitors' models anyways. So meta leaders at the meta superintelligence lab or MSL have according to the reports been discussing plugging in external models from their competitors as a near-term strategy to raise response quality while meta-continion. use in proving its in-house llama models in the long run. So a meta spokesperson said the company is taking in all of the above approach that combines partnerships with building its own systems, signaling a pragmatic focus on speed and performance, not just their own homegrown tech. So the information also reports that meta already gives employees access to Anthropics models for an internal coding assistant showing the company is actually.

actively mixing providers where it sees advantages. So meta views external model use as temporary and primarily a way to train, benchmark, and hardened Lama so it can stay competitive with Google OpenAI and Anthropic in both consumer and developer facing features. So this is one of those stories that definitely caught me off guard when we included it in our newsletter late last week. Here's why. And I've been saying this for the longest time. And I don't know if people don't believe me or what. And when, when companies are shopping around to a handful of providers and I say, no, you know, when I tell you that there's a big drop off between Google and Open AI, right? They flip, fly back and forth. You know, I'm always one A, one B. You know, one A one A. They're sharing the pedestal. It is a steep drop from Open AI and Google to. everyone else. So steep, in fact, that according to the report, that one of the companies vying

to be in that first tier is actually using their, or thinking of using their competitor models in their consumer-facing products. Let's think about that for a minute, not just internally. So we saw the report that they are apparently using Claude, Anthropics Claude, which excels at coding internally, but offering it to end up. end users via their meta AI assistant. So, you know, it might not be like a drop down where you're like, all right, I'm going to meta AI and I'm using chat GPT or, you know, I'm using Gemini 2.5 pro. Right.

So presumably it's just you're going to go to meta AI and just be using their Lama models or in the meta AI assistant that's in its Facebook, WhatsApp and Instagram offerings. But that's pretty telling. And if it's not telling for your company, I've been saying you're essentially choosing between Open AI and Google, right? Obviously, Microsoft co-pilot uses Open AI's model. So kind of the same thing there. But I mean, this is a pretty shocking story to me when I read this.

All right. Our next piece of AI news, Google dropping some pretty big updates to two of their different products. So first, it seems like they're kind of going after DuoLink. even though a duolingo is a smaller company, but it seems like Google with its new AI updates to its translate app is going after at least the duolingo audience or their share in the market. So Google is rolling out two AI powered features in Google Translate that aim to make language practice in live conversations faster and more practical. So the headline change is AI generated practice sessions inside the Google Translate app now adapt to your skill levels,

basic, intermediate, or advance, and let you choose to practice listening or speaking on demand. So lessons are also scenario based, such as asking about meal times and including short listening clips with world selection or speaking drills, where you repeat on-screen phrases. So right now, this feature is in beta on iOS and Android apps and invites you to share your motivation for learning to tailor your session content. So, you know, there's a just starting level that is planned, but is not yet available. So Google says it's worked with learning experts in current research and language acquisition, positioning this as a free lightweight practice option.

That is definitely going to cut into duolingo's casual use territory. So if I'm being honest here, I'm kind of shocked that it's taken this long. I think Google, obviously, they created this technology that translates languages, right? So it's a type of artificial intelligence technology that they essentially created many years ago before the chat GPT and generative AI boom. So it's kind of interesting to me that it took Google this long to realize that they were sitting on a potential gold mine with their Google Translate app. And I was actually surprised that they weren't doing more.

Because one of the big use cases with their Gemini Live and even with chat, GPD's advanced voice mode is learning new languages. And, you know, it's very different when you're using, you know, a kind of like a duolingo type app. Where previously, before all these companies started to try to embed AI assistance, really all that did was tell you if you pronounced words or phrases correctly. So for people looking to learn a new language, whether it's for personal or business reasons, I think this update from Google is huge. Yeah, I was not expecting Google to actually go after two different kind of startup or audiences.

So more on the second one here in a bit. But yeah, I'm curious, is anyone going to be out there now using Google Translate to learn a new language? Makes sense. All right. our next piece of AI news more meta news and probably not good one yeah multiple stories well this one is good but it stems from something that's really bad so according to reports meta is rolling out new protections over the next few weeks to stop teen users and its AI chatbot from discussing with each other self-harm suicide and disordered eating with meta AI on instagram and

Facebook with changes initially first applying to English-speaking countries. So the company says it has added guard rails that train its AI not to engage with teens on these topics and to redirect them to experts resources aiming to prevent harmful coaching or guidance. And if you're wondering like, okay, what is this update to? Well, yeah, Matt has been in some huge hot water. recently. So you can go listen to episode 592 if you want to. But essentially, reports came out that said that meta had an internal policy that actually encouraged and

trained its AI to have romantic conversations with minors. Yes, that's enough to make you throw up in your stomach. But now, after a ridiculous amount of public bash backlash. Apparently, META is working to remedy some of these disgusting things that it had baked into its AI assistant. So, META will also reportedly temporarily block teens from accessing user-made chatbot characters that could host inappropriate or sexualized conversations, limiting them to a select group of meted AI characters for now. So, like I said, go listen to 592.

But if you want the short, this move follows a series of troubling reports. So it was writers that cited an internal meta policy suggesting chatbots could have sensual conversations with minors, language that meta called an error. And they removed. But it was definitely in their policy. And it was reported that at least five different leaders and different departments approved the language. So seems like it was one of those things that, uh, it's an error. because they got caught and yeah, gross. And lawmakers are also watching this closely.

So U.S. Senator Josh Howley has announced plans to investigate Mata's handling of teen interactions. There is a deadline that he set for about two and a half weeks from now. So we'll see if Mata does cough up any internal documents that the senator requested, as well as Texas Attorney General Ken Paxson has indicated interest in probing whether the company misled children about mental health claims tied to its chat bot. All right, Marie here says episode 592 was a rough one to listen to. Yeah, it was not a fun thing to read Meta's internal policy that literally outlining how they wanted it to have romantic conversations with minors.

It was disgusting, but it was an actually reportedly approved internal policy at Meta. So about time. should have happened immediately. First of all, matters never should have had that policy. I hope whoever approved those things is no longer working at the company. And I hope that meta has an internal shift, right? Obviously, their intentions are probably a little bit different than everyone else's. They want their social network to be as sticky as possible with their meta AI assistant,

where to his credit, at least publicly, you know, Open AI CEO, Sam Altman has said, he doesn't want people spending more time using chat GPT, right? Literally he came out and said that. He wants people probably getting their answers faster and moving on. So one to definitely keep an eye on. All right. Next piece of AI news that matters this week. Salesforce is cutting 4,000 jobs due to what they say, AI agents.

So Salesforce CEO Mark Benoit said in a recent interview that AI agents allowed him to reduce customer support headcount at Salesforce from about 9,000 to 5,000 employees highlighting a rapid shift in how customer service work could be done. So according to Benoit, AI agents now handle about half of Salesforce's customer conversations with humans covering the other 50%, a drastic change from last year's reported mix. So the CEO said, AI agents are also tackling sales follow-ups, calling back more than a hundred million previously untouched leads accumulated over 26 years, which he framed as a major productivity unlock. So he described a system with an Omni-Channel supervisor coordinating when AI should handle a task

and when humans should take over similar to a driver stepping in a car in a car's self-driving mode. when it engages. So Salesforce reported 76,453 total employees across divisions as of January. So the support staff reduction represents a notable reshaping within a single function rather than a company-wide cut. So not surprising here because Salesforce's CEO, this isn't the first time. And I'm guessing it's not the last time that he kind of made these large claims about, hey, how AI is doing a big bulk of, you know, certain operations within their company. And they're really trying to push their agent force offering. It hasn't worked. And I've been obviously very honest about my approach to Salesforce and their tactics and the effectiveness or ineffectiveness of their AI and agent force.

I've had a lot of current and former Salesforce employees reach out to me and say, yeah, all this stuff you're seeing reported, it's a bunch of BS. And I think it's also telling, right, you can't look at companies stock and make every, you know, make your decision based on this. But it's pretty telling when Salesforce's stock year to date is down 22%. When you look at all the other big companies that are up 20%, up 30%, up 50%, up 50%, up 50%, or more. So Salesforce really wants to ride the coattails of this agentic success. I think their rollout was terrible. I think how they positioned agent force and even how they were charging for it originally, bad.

So essentially, I wouldn't expect this to be the last time. And Benoit, you know, early on was trying to, you know, pick fights with Microsoft. Probably not a good idea, right? Really trying to insert Salesforce and agent force into the conversation as a mayor, as a major player. in the agentic AI race hasn't worked yet. Here, another attempt. I think this is one of those things, you know, when a big CEO says, oh, we use AI to cut, you know,

customer service, you know, thousands or hundreds of jobs. You know, I think what they're trying to say is, okay, use our product, right? Yeah, I don't know. Let's keep it going. 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 Premier, 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 piece of AI news might seem small. It's pretty big. It's for developers, but I think ultimately it's going to impact everyone. So OpenAI announced the general availability of its real-time API, alongside GPT real-time, a faster, more accurate speech-to-speech model aimed at

production grade voice agents. So the real time API now supports remote MCP servers, model context protocol, image inputs, and session initiation protocol for phone calling, enabling developers to plug voice agents into existing tools in telephone systems with less custom wiring. Y'all, this is huge. I know this is technically more of a developer API story, right? So this isn't something you're going to go to chat GPD and look for GPT real time. This is for companies building on top of Open AIs technology, which is actually tens of thousands of the largest companies in the world are using Open AIs technology.

So what this means is, you know, as an example, maybe your bank instead of having to talk to, you know, a support agent and waiting on hold and being able to barely hear them because there's a bunch of chatter in the background and it takes forever, you might just be able to talk to OpenAI's GPT real time and it can access, like I said, images. It can access your company's data. So pretty big here. So the model handles nuanced speech behaviors, including better alpha numeric transcription across languages, mid-sentence language switching, right? So talk about an international impact and also tone control. And it also debuted two new voices. Cedar and Martin that are exclusive to the real-time API. So unlike traditional stitching of separate speech-to-text and text-to-speech systems,

the real-time API runs a single model pipeline, which OpenAI says reduces latency and preserves conversational nuance. So what that means is a lot of these previous historically, right, and I say historically very loosely, right, over the last year or two, when we've seen a lot of these kind of AI voice models, it's actually doing a lot under the hood. So it would first kind of transcribe the audio to text and then it would kind of think in text and it would transmit it and respond back in voice. So instead, this is a strictly voice to voice.

So it is, think of it instead of having to make multiple passes in order for the model to work. Now it doesn't. So this is a pretty big deal for even non-developers and non-technical people because we're going to start to see more realistic, more neural-sounding AI agents. I've actually called a couple of places here in Chicago. And I'm like, wait, this is an AI agent, right? Answering the phone. They've been really bad, right? Because I think they're using the older technology.

So with this new technology, I do think it's going to be a pretty big. step up in not just quality, but also ultimately usefulness to the end user. And I do think, you know, speaking of what Salesforce's CEO just said, I do think that customer service is going to be one of those areas that is going to get disrupted by this technology. But I don't think the technology has been up to snuff. I do think Gemini's models are very impressive. So we'll have to see if this new real-time API, GPT, real-time, from Open AI exceeds the quality and the response time from Gemini's model, which is one of the

top in the world right now. All right. Let's keep it going. Our next piece of AI news. Yeah, more cracks in Meta's AI partnerships. So according to Tech Crunch, Meta's $14.3 billion investment in scale AI in June. June is in looking the best. So meta did invest in Aqua hired Scale AI CEO to run their MetaI division for $14.3 billion

as well as a 49% equity stake in Scale AI. But just months later, it's seemingly starting to kind of fall apart. So according to TechCrunch, they said that Ruben Meyer scales AI and former SBP, of Gen AI product in operations who joined Meta with Alexander Wang, the former scale AI CEO, and now who's leading Meta's AI efforts. So they're saying that Ruben Meyer just left Meta roughly after only two months. And that mayor told TechCrunch, he was part of the core TBD labs. So that is one of the new four divisions of Meta's super intelligence labs.

So he said that he was part of the core. TBD labs from day one and is leaving for a personal matter. So it's pretty interesting. We're seeing a lot of reports. A lot of these people that were getting million, 10 million reported $100 million contracts are apparently leaving Meta after only weeks or months at the company. So despite the investment, five different sources told TechCrunch that Meta's TBD Labs is actively using other data labeling vendors as well.

So that was one of the big reasons reportedly why Meta did Aqua Higher Scale AI was for their data labeling technology to improve the quality of their large language models. But five different sources said that Meta's TBD Labs is using Scale AIs competitors in Surge in Merker. And some researchers view Scale AIs data quality as lower than competitors, than competitors, offerings and meta did dispute inequality issues. Yeah. So a lot of news from meta this week. So not only are they reportedly looking at potentially offering either Google Gemini or OpenAI's models to their consumers via their meta AI chat bot, but even internally their big multi billion dollar acquisition of scale AI apparently isn't going as well as it should have.

So, according to the report, scale AI has been shifting from a crowd sourced low-cost workforce to recruiting high-paid domain experts via its outlier platform. But competitors like Surge and Merker originally built around expert talent from the start and have been growing more quickly. So inside meta, multiple sources described rising turbulence, new hires from open AI and scale are frustrated by big company bureaucracy and parts of meta. previously Gen AI team have seen their scope reduced. So following a lackluster Lama 4 launch in April, Meta's Mark Zuckerberg,

accelerated deals, recruited top researchers from OpenAI, Google Deep Mountain, Anthropic, and others, and also pursued acquisitions like Play AI, Wayform AI, plus a recent partnership with Mid Journey. So Meta is also pouring money into infrastructure, announcing multiple U.S. data centers, including a $50 billion Louisiana project named Hyperion. I think that's how it's pronounced as MSL targets a next generation AI model by year end. So recent departures continue to hit Meta's AI ranks with MSL researcher, Rishab Argwal, product director Chaya Nyak, and research engineer Rohan Varma,

among those reportedly leaving after those big, you know, NBA, type contract. So a lot of recent questions around Meta's internal workings and questions about their stability and retention despite spending billions of dollars. So we thought that and we, it seemed like collectively people thought that, well, Zuckerberg and Mata and team were just going to outspend their way after, you know, Mata's Lama 4. didn't come off that great. Essentially, Zuckerberg went out and spent billions of dollars hiring the top researchers, engineers from every single big company. And it's, if I'm being honest, kind of shocking to see a lot of them leave already after only months or weeks. So that means

that they are likely leaving tens of millions of dollars on the table. So you have to be wondering why these researchers, when they, They have all the compute, right? A lot of times it's like, oh, you know, I'm going to leave here so I can have, you know, more autonomy and work on bigger projects and, you know, have more compute to do research. And so you reportedly are getting all that at meta. You're getting millions and millions of dollars or in some cases, tens of millions or hundreds of millions of dollars yet these people are reportedly leaving.

So you have to understand or think why. I'm not sure. I have thoughts on that, but we'll save that for another episode. Our next piece of AI news, pretty big news for publishers. So perplexity is launching Comets Plus, a $5 per month subscription that will share 80% of the product revenue with participating publishers. This is according to the Wall Street Journal. So the new Comet Plus plan aims to pay publishers when Perplexity's AI assistance rely on their articles to answer user questions, addressing both indexes. traffic and knew what's called agent traffic from its comet agent in the comet browser.

So perplexity says agent traffic includes actions like scanning a user's calendar and suggesting relevant articles, uh, interactions that typically bypass on site ads and page visits and aren't covered by many existing publisher deals. So the initial payouts will draw from a 42 plus million dollar revenue pool that perplexity says will expand as comment plus sign up grows. Existing pro and max subscribers are automatically contributing. That's an important one. So if you're already paying for a paid perplexity plan, you will get access to this Comet Plus.

You do have to be using the Comet browser. And the portion of your monthly subscription is being automatically contributed because Comet Plus is included in those other paid plans. So publishers receive 80% of comet plus revenue while 20% goes to compute costs for perplexity. But critics note that 80% of $5 is $4, potentially unlocking broad content access that many outlets currently price at $20 to $30 a month. So essentially publishers are saying, I don't know if that's going to really be enough

in the long run. And perplexity's current publisher program, which, includes time and fortune and some other big publishers, shares ad revenue based on traffic that AI summaries might divert. But it is unclear whether Comet Plus replaces or sits alongside this current program. So the Wall Street Journal reports that perplexity will pay publishers each time their content is used by its assistance, but details on how usage is tracked, attributed, and audited have not been fully disclosed. So, uh, engaging. its report notes the move follows past allegations that perplexity plagiarized articles,

bypassed paywalls, scraped company websites when they asked the bots not to scrape. And it's raising the stakes for a model that depends on publisher cooperation to legitimize agent-driven browsing. Like I've been saying for years, this is bound to happen. There's three things big publishers can do. They can sue the big AI model makers. They can partner with them in licensing deals such as this or they can go out of business. So I do actually like this move from perplexity.

They are the first one to at least publicly put this out there. So as a former journalist, I like this. This is maybe something that could keep some of those medium-sized publishers in business longer because it is bleak. It is bleak out there. So many of the biggest online only news publications are reportedly losing 30 to 70% of their traffic, which means they're losing a big chunk of their revenue. So if we want a future internet that is not just AI regurgitated slop, you have to root and hope programs like this stick in that big AI companies,

which a lot of the bigger companies are, you know, Google, Open AI, Microsoft, they are putting out different grants and programs. to support journalism and nonprofits putting out information. So hopefully these things catch on and stick because we need original writing. We need original reporting because that's what all these models are trained on. So it's kind of this weird juxtaposition that the AI models by essentially just ingesting the entirety of the internet. They're all training on the same internet. So by doing that, they're cutting off.

the hand that feeds them, right? In all of these newer, agentic browsers, deep research modes, right? They're taking all this high quality information that humans at some point spent a lot of time creating. But now, if those humans, if you're taking away their ad revenue, if you're taking away their subscription revenue, how are you going to feed this beast that needs higher quality data? You can't.

So this is, I would say, a red siren alert to the, uh, because I know we have people listening, obviously, to the podcast from all of the big companies. All of the big AI labs need to be doing this. Otherwise, it is going to get harder and harder. When you talk about pre-trading, reinforcement, learning with human feedback, you need higher quality data sources. And people don't know this. Before chat GPT, the GPT technology and AI technology was being used widely on the internet. So I will say probably since 2021.

a good chunk of the internet has been AI sloth, especially everything that's been posted online past 2023. So when we talk about the future of good, clean, high quality, accurate factual data, we need more partnership programs like this to keep high quality journalism, human-led journalism in place if you want future models to make sense. All right. Another big move from Google that might seem. small and is going after again a startup audience so google has added pre-made ai avatars and new editing

features to its google vids workspace video tool accelerating how teams create trainings on boardings and demo videos uh and a lot of these features are rolling out for free which is huge with some of the more premium features just for paid jemini users So the new update matters because cheaper, faster AI video creation could disrupt jobs for video editors, camera operators, and commercial actors, while also putting more powerful production tools in the hands of smaller teams and startups. Yeah, this is one of those weird ones. When Google does this, it matters, right? Yes, there's obviously dozens of very good AI avatar and AI. video editing startups. But when Google does it, it's going to be potentially disruptive.

So this, if you work in video production, you got to keep your eyes on this. So that's the bad part. I do think this is going to impact a lot of these smaller companies. They might just find trouble getting other clients because if you're a smaller business, you might just start using these new Google Vids features. On the upside, it's going to help smaller businesses tell better and more compelling stories. So this is one of those as a former video editor myself. I've spent thousands of hours in the past editing videos. So this is again, one of those ones that's kind of personal to me. I'm like, man, this is actually going to take away so many entry video, you know, video editing, video production jobs, you know, voiceover jobs. But at the same time, it's going to help so many

companies grow and provide better information to their employees and their customers. So here's What's new, users can write a script, select an AI avatar, and generate a complete video. And teams can also collaborate live in the Vids app, much like a Google Doc or Cheats. So the new features are available to Google Workspace Business and Enterprise customers and to subscribers of Google AI Pro or Ultra, but also a basic non-AI Vids editor is available for free for all users. So some of the more basic features are available for free, but a lot of these more premium ones are only for paid subscribers.

So vids now includes noise cancellation, background options, filters, appearance tweaks, and a transcript trim tool that automatically removes filler words and awkward pauses, cutting the kind of tedious edits that used to take professionals hours. So for those that prefer not to use AI avatars, you can also record yourself and let AI improve the lighting, sound, and pacing. So Google executives said that bids can fix talking too fast or using too many fillers in a few clicks. So Google says vids is aimed at teams without agency budgets and larger companies that want fast, first drafts before bringing in professional actors or agencies.

So pretty big news here. And this one, I mean, it goes after a lot. of the AI avatar companies, right? So like, Hey Jen, some of those, but also companies like Descript. So a lot of these bigger AI startups that have been around for many years that allow you to, you know, upload your videos, your meeting transcripts and cut it down, right? Instantly. And, you know, just using a transcript to edit the video as well.

So two pretty big moves I wasn't expecting from Google, kind of going after specific startup markets, but going pretty hard, right? Because from what I've seen, already, these updates to Google Vids, I think put it in the top tier versus these companies that have been trying to work and improve their products for multiple years. So it's something you have to pay attention to. All right. Next, AI News. A new study from Stanford finds how AI is cutting into entry-level jobs, but actually boosting senior roles. So the new paper from Stanford finds that since late 2020, entry-level employment has fallen sharply in AI-exposed fields, underscoring how Gen-AI adoption is reshaping who gets higher, promoted, or laid off right now. So according to

the Stanford economists who put the study together, early career workers in AI-exposed occupations saw a 13% decline in enrollment compared to 2022, compared with more experienced peers and less exposed sectors. So the biggest hits are in software engineering and customer service, where entry level roles dropped about 20% between late 2022 and July 2025, while employment for older workers in the same jobs actually increased. So overall workers aged 22 to 25. in the most AI exposed fields experience a 6% employment decline versus a 6 to 9% rise for older workers. So the researchers used payroll data from ADP for the study covering employers with a combined 25 million workers,

allowing precise tracking by age and job title, an unusually robust data set for such a recent technology shift. Also, can we just call out Stanford? Good job. This is how you do a study. You look at 25 million, you know, a pool of millions of people, not MIT that puts out a marketing study by talking to 52 people and going with their vibes. Sorry, I had to put that in there. These, y'all, certain studies are worth paying attention to.

Certain studies aren't. This is one, yes, it's a troubling study, but it also shows you kind of this. duality of how AI is actually impacting jobs. All right. And our last big piece of AI news for the week, Google has gone bananas and the internet is going bananas for Google's nano banana. So Google has released their new Gemini 2.5 flash image, which beforehand was called nanobanana.

And it was just extremely well received. I would say this is probably a top five kind of quote unquote viral AI product since chat GBT. So when you look at the LM Arena, it came in at a hundred and seventy points over the next best model for its ability to edit and create images with just text. And that is the biggest upgrade ever in terms of points on the LM Arena leaderboard. So what is this new Gemini 2.5 flash image? Well, number one, that's a mouthful to say. So I do like that Google has already just ran with and updated. Because when this came out last week, a lot of people were like,

oh, you should have just called it nanobanana. So Google has actually been putting, you know, a little banana icon when you choose it. I believe in AI studio, it says the word nanobanana now versus if you just have to be like, okay, is Gemini 2.5 flash image? Is that the right one? But here's what's different in what's very impressive about this model. Well, one, it is a multimodal by default. And that is huge because Google does have their other AI image model.

Imagine 4, which is extremely incredible and useful. However, this is different because it allows you because it is multimodal by default and it has an understanding because it's part of Gemini 2.5. It understands how the world works. it understands relationships between objects. So when you are editing in natural language, so you can upload a photo and edit it with your words, right? It's important for it to be multimodal by default.

Also, the character consistency off the charts, right? So a lot of people when the Open AI GPT40 image gen kind of went viral, they noticed, okay, well, when I upload a photo of myself as an example, it always changes certain features, right? I think Open AI did that by default. They probably did it for safety and guardrails, but Nanobanana didn't, right? They're allowing essentially pinpoint accurate character consistency. So you can upload a photo of yourself and then essentially do whatever you want, right?

But keep that character consistency. You can upload, you know, 10 different objects and then Google's new model can combine them all in one photo. and then you can keep that character consistency throughout. So this is huge, a huge jump in quality and even capabilities just for all creative fields. So pricing and access are structured for scale. So Google AI Studio currently offers about 100 free daily edits for users, while the API usage is extremely cheap at just under four cents in image, which is a fraction of what the other AI text image

or AI image editing models charge that are not nearly as powerful. So like I said, some of these standout features of the new nanobanana Gemini 2.5 Flash image is the ability to talk in natural language, character consistency, and just being able to iterate with different images. So pretty big news.

And I think this isn't the last you've heard on this show of Mano Banana. We're probably going to be having a few more episodes covering it. All right. That's not it. There's a whole lot more. Now we're here for the wrap up section of what's new and what's next. Because recently there's too much to cover for our big news stories. So here's a quick bullet point rundown of some other noteworthy.

updates that maybe didn't make our top stories list. All right. Here we go. Rapid fire. So Gemini's C-L-I, their code tool, their coding tool for the desktop for the code line editor is now integrated into the editor Z. Anthropic, pretty exciting one here. They're piloting a new Claude for Chrome feature, only rolled it out to a thousand beta

testers. But this is pretty cool. It's an agentic, it's an agentic. Um, it's an agentic, uh, use case that actually uses your browser and not a virtual environment. Microsoft has released its two kind of flagship models, uh, MAI voice one and MAI one preview. Uh, Anthropic will start training its chats on user data, uh, after not doing that for the

first couple of years. Uh, notebook L.M has teased flashcards literally the day after I said, oh, they don't have cards, but they need them. Notebook L.M is now teasing those. They may be rolling out soon. XAI unveiled their GROC code fast one coding model. Some more notebook LM rumors. They may be adding new voices and new modes soon.

Microsoft co-pilot rolled out a study and learn option, very similar to those features found in chat, GBT, and Gemini. OpenAI release some major code. updates that have been extremely well received. I'm actually surprising kind of. I'm surprised the conversation around Claude code has been getting more and more negative over the last few weeks.

And this new codex update, it's been very well received. So we'll probably be talking to that about that more in the future. Chat GPT may soon release conversation branching, which could be big. Chat GPT team is no more. That plan has now been renamed chat. business. According to Tabor on Twitter, OpenAI may be releasing a chat GPT health team plan soon, which includes HIPAA compliance. That's pretty big. And chat GBT may be updating its thinking mode with light standard extended and max thinking options for paid users. That was a ton of AI news.

Let's quickly recap the main stories. So we started with talking about meta, potentially using Google Gemini or Open AI's models to help their own Lama models along. Google translate with some huge AI updates and kind of essentially challenging duo lingo in their audience. Meta has tightened their safeguards on its AI and how it interacts with teen users. Salesforce reportedly cut 4,000 jobs in customer service due to AI agents. OpenAI launched their real-time API called GPT RealTime, which I personally think is going to be huge. Metas, a scale AI investment is reportedly showing some early strain with some problems and people leaving. Perplexity has unveiled their new Comet Plus that will pay publishers.

Google, their bids product has added AI avatars and easier and speedier video production. A new study from Stanford showed that AI is actually hurting entry-level jobs for younger workers, while older workers are actually finding more jobs. And last but not least, Google released their mega viral, extremely capable Gemini 2.5 Flash Image, nano-banana model. That was a ton to go through. I hope this show was helpful. If so, please repost this.

We normally do this every Monday with the holiday. It was just the AI news that matters on. But for the most part, here's our lineup on Monday. We bring you the AI news that matters. Most Tuesdays, we bring you a hot take Tuesday, our hot take on something going on in the world of AI. On Wednesdays, we give you practical use cases with AI at work on Wednesday, showing you how to use different tools and modes and giving you practical use cases.

And for the most part, on Thursday and Fridays, we do different interviews with the world's leading experts. That's going to help you accomplish your goals in being the smartest person in AI at your company. then you got to go to our website, your everyday AI.com. Go sign up for the free daily newsletter. Thanks for tuning in.

We'll 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.