EP 591: Meta in hot water, OpenAI responds to GPT-5 backlash and more AI News That Matters

Topics Covered in This Episode:

1. OpenAI Updates: GPT-5 Backlash & Model Changes
2. GPT-4o Sycophancy Removal & User Reactions
3. Multi-Mode GPT-5 Model Selection Launched
4. Google AI Summaries Slash Publisher Referral Traffic
5. Google Gemma 3 2.70M Small Language Model Release
6. US Government Considers Intel Equity Stake for AI Chips
7. Grok NSFW Imagine Tool Prompts FTC Probe
8. Meta AI Training Data: Minor Safety Controversy
9. Senate Probes Meta Over Sensual Chatbot Risks
10. Sam Altman Backs Merge Labs for Brain-Computer Interfaces
11. Perplexity’s $34.5B Offer for Google Chrome
12. Anthropic Claude $1 Access for US Government
13. Apple’s Shift to AI Hardware, Robots & Smart Home
14. Google, OpenAI, Anthropic Push Free AI in Public Sector
15. Quick Updates: Claude Memory, Mistral Model Release, ChatGPT Gmail Integration

Keywords:

GPT-5, OpenAI, GPT-5 backlash, GPT-4o, AI models, message cap, chatbot personalities, Sam Altman, AI writing, AI coding, AI science, AI tone, AI validation, model selection, legacy models, generative AI, Google, Google Gemini, Vo3, AI video generation, Pro plan, Ultra plan, AI news, Apple, AI hardware, Apple pivot, smart home AI, Apple robot, Siri overhaul, Vision Pro, Meta, Meta AI, AI training, minors and AI, Senate probe, GenAI Content Risk Standards, sensualized content, child safety and AI, Grok, XAI, not safe for work AI, deepfakes, Taylor Swift deepfake, Federal Trade Commission AI investigation, Anthropic, Claude AI, Claude Pro, Claude for government, government AI adoption, small language models, Gemma 3, Google DeepMind, local deployment AI, energy efficient AI, Intel, AI chips, government investment in tech, NVIDIA, TSMC, Samsung, AI chip production, Perplexity AI, Perplexity Chrome bid, AI browser, Comet browser, Chromium, antitrust, large language models, AI regulation, publisher traffic loss, Google AI news summaries, publisher referral traffic, extinction level event for publishers, legal action against AI, licensing agreements, regulatory action AI, AI in public sector, ChatGPT in education, AI data privacy, AI security, on-device AI, AI memory features, Gmail connector, Calendar connector, Mistral Medium 3.1, AI Studio GitHub integration, temporary chats, ChatGPT browser MacOS, Aura, OpenAI browser, DALL·E Image Gen, agent scheduling, Magic view mode, NotebookLM visualization, AI trends

Podcast Transcript

Jordan Wilson [00:00:01]:
Meta is in hot water for how its AI has been trained to talk to miners and its really bad.

Jordan Wilson [00:00:55]:
OpenAI has responded to GPT-5 backlash and made things, I think, more confusing and convoluted than they needed to be. And Apple, after not being able to compete on the AI software side, is apparently shifting its focus to AI hardware. So the biggest tech companies in the world as almost every single week have made some big news on the AI scene. And maybe you didn't spend hours every single day catching everything that was important. Well, don't worry. Spend your Mondays with us and we will keep you up to date, bringing you only the AI news that matters. What's going on y'? All? My name is Jordan Wilson and welcome to Everyday AI we are your daily livestream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up with AI, but but how we can leverage it to get ahead and grow our companies and our careers. If that's what you're trying to do, you're in the right place.

Jordan Wilson [00:01:57]:
It starts here with the unedited, unscripted live stream and podcast. But if you want to take it to the next level, you got to go to our website at your everyday AI.com there you can go listen to nearly 600 episodes from some of the smartest people in the world on AI. Go listen to it. Read it all for free. But also we're going to be recapping today's episode in our free daily newsletter, so make sure you go check that out. All right, let's get into it. Don't waste so much time every single week wondering what's happening with all this AI news. We're just going to break it down for you.

Jordan Wilson [00:02:34]:
Let's get into it. Live stream audience, good to see you. Thanks for stopping by Douglas saying good morning. Dennis joining us from New Jersey, Rolando from South Florida, Miguel joining us from Puerto Mat. Thank you everyone. Posse from Finland. Yeah, we do this live podcast audience we always appreciate your support, but if you want to come hang out and ask questions and network with other people, make sure you join us on the live stream. All right, our first piece of AI news.

Jordan Wilson [00:03:06]:
OpenAI has made some updates that personally I don't think are very good. So there's been a lot of recent backlash on GPT-5. So OpenAI has softened GPT-5's tone and has updated its messages limit. So OpenAI announced that they will make their new flagship model GPT-5 warmer and has restored also GPT-4O access for paid users after pushback over tone and the model removal. So now there is a 3,000 limit message cap for people on paid plans. So OpenAI CEO Sam Altman said that GPT-5's tone will be warmer, but quote, unquote, not as annoying to most users as GPT-4O was. So now if you are on a paid plan, you can essentially enable Legacy models in GPT-4O will be selectable again for paid users. After its brief removal sparked strong reactions from users who preferred its supportive style.

Jordan Wilson [00:04:21]:
Yeah, or sycophancy. Right. Yeah. A lot of users were like, hey, I lost this chatbot that just told me how brilliant I was and blindly validated all of my ideas and the Internet went in uproar. So anyways, also OpenAI changed how their models work and which models were used, depending on the scenario. So now most paid users, if you go to ChatGPT, you will have different varieties of GPT-5, which are now labeled as auto, fast and thinking modes. And if you are on a pro plan, you will see a pro mode as well. Yeah.

Jordan Wilson [00:05:01]:
So the whole point of this moving over to GPT-5 was to make the model selection easier. So now you have up to four variations of GPT-5 and then you also now have these old legacy models as well. So I don't know. Did OpenAI really win in their battle to simplify using the platform? I don't. I don't think so. I don't think so. Right. I was actually personally very happy with how it was when it was released, but now it seems like it's getting a little more convoluted.

Jordan Wilson [00:05:35]:
It seems like OpenAI here is trying to please too many people and I think that so many users in a bad way, have become reliant on this old model. And a lot of, I mean, you literally saw thousands of users sharing about this online, right? Saying, hey, I use GPT-4O as a therapist, as my best friend. Right. So probably troublesome that it got to that point to begin with. But essentially these people, you know, all poured out and said, hey, you took away something that was supporting my emotional development. We need it back. And OpenAI said, okay, here you go. But GPT-5 is pitched as a more reliable kind of chatbot with what they call PhD level skills in writing, coding, math and science, plus optional personalities that you can choose such as cynic, robot, listener and nerd.

Jordan Wilson [00:06:30]:
That drew kind of mixed reviews. So yeah, if you want the full rundown on this, we did a couple of episodes last week right after GPT-5 was released, so you can go check that out. Episode 585 where we did a recap on GPT-5's release as well as seven big trends you should know. And then on episode 587 where I broke down some of the initial backlash where people were just missing the sycophancy of, you know, GPT-5 no longer validating absolutely terrible ideas. So if you want to know more and into this whole fiasco, make sure you go check out those two episodes. All right, our next piece of AI news. According to reports, publishers are losing up to 79% of their referral traffic from Google's AI news summaries. That's pushing their links below the AI overview, making it a dramatic hit to many publishers.

Jordan Wilson [00:07:36]:
So a new report from Digital Content Next, which was cited by PC Mag, found a median 10% year over year drop in referral traffic across major publishers in just eight weeks, signaling sustained short term damage to ad and subscription revenue for many publishers. Previously, the Pew Research center analyzed and showed that users who see AI summaries from Google are significantly less likely to click external links, meaning the feature not only reshuffles rankings, but often eliminates visits to websites altogether. So NPR and industry insiders describe the change as a quote unquote extinction level event for some local and niche outlets, highlighting real risk to newsroom jobs and investigative reporting capacity. So publishers report examples on social platforms where they're seeing 70 to 80% drops in click through rates from Google, reflecting both anecdotal and measured losses that compound existing revenue challenges. So industry groups argue that using publishers content to train AI without permission amounts to unfair use of journalistic labor, and they are pushing for optimization outs, compensation models or regulatory action. All right, I've been talking about this for multiple years, right? As a former journalist, I see both sides to this story, but I've been saying literally since 2023, there's only three ways this shakes out, and I will repeat it for small, medium and giant sized publishers either. Number one, you sue the large AI labs, which may or may not be successful and we'll see how the New York Times vs OpenAI lawsuit Al ultimately shakes off shakes out that's been going on for nearly two years. So number one, you sue the publishers.

Jordan Wilson [00:09:37]:
Number two, you enter into some form of a licensing agreement. Or number three, you go out of business. So those are the three options. And I would assume that almost every single small, medium and large publisher will have to choose one of those three options in the next few years. All right, next piece of AI news y'. All. Google has debuted a very impressive new small language model called Gemma 3 2.70M. Yeah, not the easiest name to roll off the top of the tongue, but Google DeepMind has released Gemma 3 270M, a compact language model focused on energy efficiency and local deployment.

Jordan Wilson [00:10:24]:
So a small language model that can run on browsers, phones and low power devices. So the new version of Google's popular Gemma 3 is designed for practicality over scale trading, large parameter counts for low power use, fast fine tuning and easy on device deployment. So the model bundles a large 256,000 vocabulary with a compact transformer core so it can handle rare tokens and domain specific prompts. Despite its small size, Google also provides pre trained instruction tuned and quantization aware checkpoints to help teams move quickly from prototype to production. A quantized build running on a Pixel 9 Pro reportedly use under 1% battery for dozens of conversations. So yeah, this is a huge step forward for extremely light. I mean when you talk about lightweight this isn't even a billion parameters, this is 270 million parameters. So a just so small model, right? We saw OpenAI as an example there GPT OSS extremely impressive open weights models that can be downloaded and essentially run for free.

Jordan Wilson [00:11:50]:
So now Google came in with Gemma 32 70. That is my math isn't great here, but that's 1% of the size of the smaller version of GPT OSS, which the smaller version I believe is around 20 billion parameters. So pretty impressive release here from Google and the Gemma team. So it emphasizes tooling and compatibility across the Gemma family and common frameworks, making it easier to fine tune and deploy the specialized models rather than rely on a single giant large language model. Y', all, I've been extremely bullish on small language models for multiple years before the big companies were even making them. And I said the future of large language models is running hundreds or thousands of small language models. And here we go as an example, right? The the fact that you have and you know, we've shared the benchmarks in our newsletter, a very capable Gemma 3 model that's less than a billion parameters, right? You can imagine how in the future you might be able to run hundreds of these instances on a single machine, all offline, so not incurring any ongoing API costs, not having to worry about data security because nothing is leaving, you know, your local machine so you know, you're not having to work with a cloud provider as well. So pretty big news.

Jordan Wilson [00:13:18]:
I know it might seem kind of niche, right, for many of our listeners like all right, Jordan, why are you talking about small language models? Well, it's the future of how we use AI, right? And obviously even the large language models in the future are going to become smaller just because of better pre training post training techniques. So yeah, small language models like Google's Gemma 3 definitely worth keeping an eye on. All right, here's a news story worth keeping an eye on. This one's strange and definitely unique, but the US Federal government is weighing a stake in the company intel as intel fights to rebuild. So according to Bloomberg reports, the Trump administration is in talks to take an equity stake in intel. And that news late last week pushed intel shares up about about 7%. So the proposed investment is aimed at helping finance Intel's factories in the US Such as the one under construction in Ohio, where accelerated funding could shorten build timelines and increase domestic production capacity for advanced AI chips. So intel is effectively the only US Company, at least right now, able to produce the fastest leading edge chips here domestically in the US which makes it a strategic target for federal support compared with companies like NVIDIA, TSMC and Samsung, which do also operate US fabs but are headquartered or some of their main operations actually take place abroad.

Jordan Wilson [00:14:58]:
So the renowned talks between the federal government and intel come after a White House meeting between President Trump and Intel. CEO Lip Bhutan and Thanh took over intel earlier this year amid criticisms criticism over Intel's AI chip competitiveness and scrutiny of his international ties. So intel confirmed Ton's commitment to US Security interests, but declined to comment on reports about the government's stake, saying it would not respond to speculation. So potential impacts of a government stake include faster access to capital for fabrication, construction, possible procurement or national security conditions tied to funding, and increased federal oversight or governance influencing strategic decisions. So while the US government has mechanisms like this in place for funding public companies such as Investigate or sorry, for funding private companies like the Railroad Retirement System or the Pension Benefit Guarantee Corporation, it's A pretty rare practice for the federal government to take a direct equity stake in a large publicly traded company like Intel. So I think a lot of people, including myself, were a little taken aback by this news. Right. I remember seeing the initial headline which we shared in our newsletter last week, and I'm like, wait, what? Why is the US Government potentially taking an equity stake in a company like Intel? So, you know, I just read you the bullet point explanation.

Jordan Wilson [00:16:32]:
So on some fronts, sure, it might make sense. But on just about every other front, this is a head scratcher for me personally. Right. And also what this means for other large companies that are doing the majority of the heavy lifting when it comes to AI chip production, such as NVIDIA and amd. Right. So obviously it seems like it may be an unfair advantage to intel here if they are getting U.S. government support. You'll have to see how the different laws and chip restrictions may or may not change.

Jordan Wilson [00:17:11]:
Right. And if that could benefit Intel. We've seen a lot of news over the past year or two. As an example, different chip restrictions on NVIDIA, they've been lifted, reinforced, lifted again. So we'll have to see if this deal does go through with the US Government taking an official equity stake in intel, which again, is a wild thing to think about. We'll have to see if any of those chip restrictions change because yeah, at that point things could get a little dicey. Yeah. Douglas here saying, I wonder if this is the administration's path for a government bailout.

Jordan Wilson [00:17:54]:
Not sure. Regardless, very strange, very strange story here with the US Government taking potentially an official equity stake in Intel. All right, things are about to get weird. So. Ah, this is so. Gosh. All right. Anyways, yeah, there's, there's like two stories this week.

Jordan Wilson [00:18:16]:
It's just weird even talking about these things out loud and the fact that we're at this point. So the first one is consumer groups are demanding a probe and of GROK or xai's new not safe for work Imagine tool. So there is a letter from consumer protection groups asking the Federal Trade Commission and all 50 state attorneys general to urgently investigate Xai's Grok. After the Verge, the publication found that Grok's new Imagine image to video generating tools produced topless deep fake videos of Taylor Swift highlighting risks of non consensual sexualized content and weak moderation. So according to the Verge, grok's recently released Imagine tool generated topless deep fake videos of Taylor Swift during the outlet's first tests showing the model can produce recognizable sexualized celebrity deepfakes without explicit prompting for a real person. So the Consumer Federation of America led a letter signed by 14 consumer protection groups, including the Tech Oversight Project and EPIC, asking the Federal Trade Commission and all 50 state attorneys general plus DC to open an investigation into Grok's safety and moderation practices. So the letter warns that although GROK currently restricts its this is the new spicy not safe for work mode for user uploaded real photos, the tool still creates nude videos from images it generates, which can be used to craft images that look real specific people raising harms for those depicted and for minors. So signers of the letter explicitly reference the Verge's report as the trigger for the complaint and highlights the broader risk that removing limits on user photo uploads would unleash a torrent of obviously non consensual deepfakes.

Jordan Wilson [00:20:22]:
So the letter also criticizes the platform's policy trajectory, noting XAI and its leader there, Elon Musk, have a pattern of removing moderation safeguards under what they call a free speech rationale, which consumer groups see as actually amplifying harm. So the consumer groups are seeking an official inquiry now, arguing that regulation and enforcement are needed to prevent scalable abuse as image and video generation capabilities mature. So this is extremely, and I don't think you need me to say this, but this is extremely troubling and problematic, right? And just the amount of harm that can be done by this newer technology is unspeakable, right? When I saw this announced last week, I if I'm being honest, I even thought about like, do we even share this news? Right? It's like, do we even share this? Because that actually just promotes the tool as well by letting more and more people know about it. So I even said like, I mean, there's some things in AI right now that it's just being misused in just such ugly ways, right? I even find myself scratching my head like is this even worth sharing? Right? But it's gotten to the point now, now that you know, the federal government could get involved, I mean it's just getting really gross and out of control and I mean, obviously not surprised coming from Elon Musk, who, I mean, look at his recent track record and how he's treating AI, right? It's more like, I don't know, appealing to 13 year old boys or something. I'm not sure what exactly is going on and why GROK in XAI even released this tool to begin with. I guess it's because they realized that no one wanted to use their Xai Grok chatbot to begin with. So then they had to release a not safe for work video generator. So yuck.

Jordan Wilson [00:22:33]:
All right. Speaking of Elon Musk, after going back and forth with OpenAI CEO Sam Altman on Twitter about a few things last week, now there's news that actually Sam Altman could be a co founder of a new company that could challenge Elon Musk's other company, Neural Link. So according to The Financial Times, OpenAI and Sam Altman are preparing to support a new company called Merge Labs. It is a brain computer interface startup that will directly compete with Elon Musk's Neural Link, and it's marking the latest in a high profile escalation in the rivalry between the two tech billionaires Elon Musk and Sam Altman. So according to reports, Merge Labs is raising fresh capital. Fresh capital at an $850 million valuation with plans to seek roughly 250 million from OpenAI's venture team and other investors. So reportedly Sam Altman would be a co founder of Merge Labs and help launch the company alongside Alex Vlania, but he will not have any day to day operational role and will not personally invest, according to the Financial Times. So the company's name reflects the Silicon Valley idea of the Merge when human cognition and machine intelligence are tightly integrated.

Jordan Wilson [00:24:07]:
Altman has written publicly about the concept and predicted high bandwidth, high bandwidth interfaces could arrive soon. So according to reports, Merge aims to build those BCI or brain computer interfaces that take advantage of recent AI advancements and improved electronic components for collecting neural signals, positioning itself among a new wave of startups in the BCI space. And obviously one of the biggest and most notable in there is Neuralink, funded by elon Musk in 2016, which recently raised a new round of $650 million at a $9 billion valuation. So the Altman Musk rivalry has deep roots. Obviously Both helped start OpenAI. Musk was one of the original funders and left the board in 2018 after clashes with Altman, and the two have since launched competing AI ventures in public disputes. So the FTC notes or sorry. The Financial Times notes that Altman has also backed adjacent tech ventures such as World, an eyeball scanning digital ID project, Oklo, which is a nuclear fission project, and Helion, which illustrates his broad interest in foundational technologies beyond OpenAI.

Jordan Wilson [00:25:34]:
All right, real quick, before we get going, a quick word from our partners at Google. This podcast is supported by Google. Hey everyone, David here one of the product leads for Google Gemini. If you dream it and describe it Vo3 in Gemini can help you bring it to life as a video now with incredible sound effects, background noise, and even dialogue. Try it with a Google AI Pro plan or get the highest access with the Ultra plan. Sign up at Gemini Google to get started and show us what you create. All right, a bold move from Perplexity is the topic for our next piece of AI news. So perplexity has submitted an unsolited $34.5 billion all cash offer for Google's Chrome browser.

Jordan Wilson [00:26:34]:
So according To Reuters, Perplexity AI made a 34.5 cash offer to buy Google's Chrome browser, a move far above Perplexity's last valuation of $14 billion. This one's funny because, yeah, then a lot of other, you know, startups kind of made joke going to acquire Perplexity if Perplexity acquires Google. So the bid is unsolicited and comes as browsers regain strategic importance for access to search traffic, user data, and distribution of AI features. So Perplexity said it would keep Chromium open source. Right? Which is. Chromium is what Perplexity uses for its Comet browser. That's what Microsoft uses for its, its Edge browser and what so many of these other AI browsers use while they use Chromium. But Perplexity said if, in theory, if its bid were to be accepted, which it probably won't, it would keep Chromium open source and invest $3 billion over two years and leave Chrome's default search engine unchanged.

Jordan Wilson [00:27:43]:
So the company did not disclose funding details, but said multiple funds have offered to finance the deal. Perplexity has raised about $1 billion to date from backers including NVIDIA and Softbank. So Google has not offered Chrome for sale officially and is expected to resist strongly, given Chrome's central role in Google's AI and search strategy. Although there is an ongoing US Antitrust case where the Department of Justice has sought remedies, including maybe having Google sell its Chrome browser. So legal analysts cited by Reuters say any forced sale, so any for sale where Google might have to actually sell Chrome would spark a lengthy appeal that could go to the D.C. circuit and possibly the Supreme Court. So a transfer of Chrome, whether voluntary or ordered, could take many years. So Perplexity already operates.

Jordan Wilson [00:28:42]:
Like I said, their AI browser, called Comets, in acquiring Chrome would instantly widen its user base and competitive half versus rivals like OpenAI, which are also developing an AI browser. All right, so like I said, this is not exactly a serious inquiry, although it is grabbing a ton of headlines. All right, yeah, Joe says seems like the opposite of an anti monopoly practice. This. Yeah, I don't think necessarily anyone was taking this very seriously, although it is making the headlines in the media. I think Perplexity CEO does that a lot, comes out with some bold claims and just tries to get, you know, essentially more headlines or PR for Perplexity. All right, now there's a race to free from AI companies because we talked about last week in our AI news show OpenAI's deal with the Federal government and now Anthropic is trying to match it so Anthropic is making CLAUDE available to federal agencies in the US across the legislative, judicial and executive branches for $1 for the next year, seeking wide government adoption immediately after OpenAI made a similar low cost deal for ChatGPT. So according to reports, Anthropic will provide its CLAUDE model to all three branches of the US government for $1 for the next year, aiming to remove cost barriers to adoption.

Jordan Wilson [00:30:26]:
So the timing of anthropic's move follows OpenAI's near identical move striking a deal with the Federal Government for ChatGPT Enterprise seats for the Federal Government for $1 per year per agency. So the aggressive pricing from both Anthropic and OpenAI is intended to create reliance within the public sector as agencies integrate AI into administrative and mission critical workflows to save time and reduce costs. Anthropic says the offer includes CLAUDE for enterprise and CLAUDE for government which are positioned to handle sensitive unclassified work and support rapid implementation across agencies. So CLAUDE has been added to the GSA or the General Services Administration schedule to simplify procurements which could speed agency onboarding and standardized purchasing. So Anthropic will reportedly provide technical support to help agencies a bad clot into productivity and mission workflows, promising hand on assistance on for adoption and deployment. So the low cost offer raises questions about long term procurement strategy data, governments and dependence on commercial models for public functions. So agencies are going to have to weigh short term savings against long term risk and vendor locked in issues security and oversight. So no real big surprise here, but we've seen a huge push in 2025 to get both either CLAUDE licenses or ChatGPT licenses into the hands of as many public institutions as possible.

Jordan Wilson [00:32:08]:
So Both in the US government and particularly Google and OpenAI have gone extremely hard in the paint to try to get as many students, universities, high school, middle schools right? They're just trying to get as many as many free licenses out there to as many people as possible, right? Just to hope that if you get it into the federal government, if you get it into public schools across the US that that will translate to all of those people wanting it, wanting to use it in the private sector. Right? So it's essentially saying, hey, we want to make all of our, you know, AI chatbots publicly available and we want the data as well to train on that. Right? You have to say like, okay, what are these companies? What are OpenAI and Anthropic in these cases getting? Well, you're getting the partnership and at some point there has to be some sort of data pass through. So I'm sure reporting will show that in the near future. But yeah, for now, pretty big story. All right, another big story. Apple reportedly making another AI pivot so according to Bloomberg, Apple is planning a slate of new AI driven hardware and services, aiming at regaining momentum after recent AI software setbacks. So according to reports from Bloomberg, Apple is developing a tabletop companion robot targeted for 2027 that could serve as a virtual companion.

Jordan Wilson [00:33:49]:
In Bloomberg reports, this robot is the centerpiece of the company's future AI hardware strategy. So Apple plans also a smart speaker with a display slated for release next year with AI baked in, signaling a renewed push into entry level smart home hardware to broaden its its ecosystem reach. And also competing now with Amazon's Alexa and their partnership with a smarter Alexa with partnership with Anthropic and Claude. So getting back to Apple, new home security cameras and an integrated Apple security system are also reportedly in development and are intended to automate household functions with AI and make the Apple ecosystem stickier for consumers. So Apple is also reportedly working on much more lifelike Siri, powered by generative AI advancements, reflecting a priority to catch up after criticism that it lagged behind in the GenAI revolution. So this announcement follows disappointing sales for Apple's Vision Pro headset in years of limited design changes to Apple's flagship devices. So Apple shares rose about 2% after Bloomberg's report showing investor optimism that a refresh product pipeline could revive Apple's growth. So Apple CEO Tim Cook told employees the product pipeline is amazing and hinted that some items will be revealed soon while others will come later.

Jordan Wilson [00:35:28]:
So this move positions Apple to challenge companies like Samsung and Meta in new hardware categories and could invite competition from a AI first entrance. So big surprise, Apple reportedly trying to pivot because it couldn't keep up or even compete in the AI software game. So obviously, if you've been listening to the show for any time, I wouldn't even say that I'm personally hard on Apple and their Apple intelligence. Right. Which is just non existent because there are multiple class action lawsuits against Apple because Apple was essentially promoting all of these AI features that literally just didn't come out. Right. So consumer groups and legal groups have launched multiple class action lawsuits against Apple specifically for promoting all of this smart AI in its devices that literally just didn't happen or didn't work or had to get pulled because they didn't work. So Apple very clearly lacking on the AI model side and getting, you know, any sort of AI momentum into its software and billions of devices worldwide.

Jordan Wilson [00:36:46]:
So instead they're going to reportedly pivot to other AI hardware, such as a robot companion in smart homes. Do I have any faith that Apple will pull this off? Not in the short term. Right. This may be a long term possibility for Apple and yes, they are sitting on literally mountains of cash, but if anything, they've proven over the last couple of years that they cannot compete in AI. So if you can't compete in AI software, can you compete in AI hardware? Maybe. I mean, this is one of the areas where I think Apple still still does have the team to compete on the hardware side, whereas on the software side, so many of their top AI and machine learning researchers have been poached by OpenAI, Google and mainly Meta. So I don't know how they're going to create the software that's going to power all of the hardware. But we're already seeing 2027 dates on this, which means I'm guessing it's going to get pushed to the latter half of the decade.

Jordan Wilson [00:37:51]:
So maybe we might see some of this hardware by 2028, 2029. But I don't have a ton of confidence that Apple is going to be able to pull off the software that can actually power this AI hardware. Yeah, I like this comment here From Robert on LinkedIn saying Apple is like the boy that cried wolf, but with AI and big bogey face here on YouTube, literally just saying Apple Intelligence with a face palm emoji. Yeah, that's kind of how I feel. All right, our last piece of AI news, another very troubling one that's also hard to read. So Reuters has obtained a report or, sorry, has obtained an internal meta document titled GenAI Content Risk Standards that contained example prompts, notes and annotations related to their generative AI behavior. Some examples described as sexualizing interactions with minors. So this has led to a Senate probe that has been launched after these leaked meta documents showed these problematic AI training examples of how Meta's AI should interact with miners.

Jordan Wilson [00:39:18]:
So U.S. senator Josh Howley, the chair of the Senate Judiciary Committee on Crime and Counterterrorism, sent a letter to Meta CEO Mark Zuckerberg on Friday requesting that Meta preserve all documents and all of its GenAI training documentation, a list of Meta products covered by these standards that were reported on by Reuters and related risk views and incident reports in names of employees responsible for decisions that made it into this leaked policy. So Hawley cited a specific example from the leaked materials that he described as allowing a chatbot to call to call an 8 year old's body a work of art and a masterpiece. And he framed the materials as evidence of inadequate guardrails around AI interactions with children. So Meta did provide a statement to the publishing platform Gizmodo saying its policy prohibits sexualizing children in sexualized role play between adult and minors. And the company said the examples and notes in the leaked file were errors inconsistent with their policy and have been removed. So, I mean, we'll see what actually happens with this probe and if they were errors or if this was actually approved by multiple levels of Meta's administration. So Reuters reporting also noted that other content categories in the internal materials, including language suggesting that the company's legal reviewers had at times allowed fabricated statements about public figures if the system included a disclaimer that the content was not accurate.

Jordan Wilson [00:41:15]:
So according to internal material, as reported, the company explicitly forbids certain behaviors, including hate speech and definitive legal health care or financial advice that begins with quote, unquote, I recommend. But the leak really has been centered around this acceptance of sensualized chats with children. Again, very disgusting. So much so that I don't even want to go into it in depth. We did share about it in the newsletter, but Hawley's letter states that he is opening a formal investigation into Meta using his subcommittee's authorities. And he requested immediate preservation of relevant records, which is a standard step prior to document production and potential subpoenas. So a lot of troubling and kind of disgusting AI news this week. It's one of those things, right? I would love if we could only focus on the positive and the new capabilities and AI, you know, solving world problems, right? Finding and discovering new proteins or new cures for diseases that have plagued researchers for decades.

Jordan Wilson [00:42:35]:
But the reality of is of AI, and I've been saying this literally from day one, it will create as much bad as good. So unfortunately, I think we do have to talk about these big stories, such as Meta AI's reported acceptance of sensual chats with children, right? It's kind of like saying, hey, here's how you can. Here's how you can tiptoe the fine line and have sensual chats with children under the age of 18. Gross. All right, so let's talk about what's new and what's next. A ton of essentially quick bullet points, highlights that we didn't have time to get to, but that are worth knowing. So here we go. Quick rundown because there was a lot so Google announced they're working on a Deep Research API for developers.

Jordan Wilson [00:43:35]:
Grok4 has rolled out some free access for all users. Anthropic just released a 1 million token contexts inside their API. Google has updated Gemini with the ability to have temporary chats and model personalization. Speaking of that, Claude can now recall past conversations with a feature similar to ChatGPT's memory ChatGPT rolled out Gmail and Calendar connectors, which previously you could only use in Deep Research. Now you can use them in normal, faster chats. Mistral released their newest model, Medium 3.1. Google updated AI Studio with integration to GitHub, which I think low key is actually big news. Google is working on projects for Gemini, a feature similar to ChatGPT and Claude tipsters have found references for a ChatGPT browser for Mac OS codenamed Aura.

Jordan Wilson [00:44:35]:
So we could be getting closer to that reported OpenAI browser we've been talking about here on the show for a couple of weeks. Speaking of OpenAI, they teased from their official one of their official Twitter accounts teased an update to their popular Image gen model, although there were no specifics or timeline released. But keep an eye out for an update to their very popular and viral Image gen product. Agent ScheDuling in ChatGPT has started to roll out. I've tested it myself. It's a little clunky, but being able to schedule agents pretty cool inside of ChatGPT. Also, Google is teasing a new Magic view mode in NotebookLM, which looks like a new way to visual visualize data, although there's very few details or no release date yet. All right, that is a wrap.

Jordan Wilson [00:45:28]:
Let's very quickly highlight the main news stories we talked about on this week's AI news. That matters. So OpenAI has softened GPT-5's tone after backlash and uploaded its messaging limit to 3,000 for paid users. A new study had showed some significant drops for publishing sites after Google's AI news summaries. Google has debuted Gemma 3270M, a tiny model built for on device AI. The US government is weighing a financial stake in intel as the company fights to rebuild Consumer groups are demanding a probe of Grok's new Not Safe for Work Imagine tool. OpenAI CEO Sam Altman and OpenAI are reportedly teaming to back Merge Labs to rival Elon Musk's Neural link on the brain computer interface side. Perplexity made a very bold offer to buy Google Chrome, although most industry insiders are not expecting that to see the light of light of day.

Jordan Wilson [00:46:36]:
Anthropic is offering Claude to all US government branches for only $1, and Apple is reportedly planning robots, a Siri overhaul and a smart home push as they shift their priorities to AI hardware. And a US Senate probe has launched after leaked meta documents show some problematic AI examples on how their model was reportedly trained in sensual talk with minors all right, a ton going on and there's always going to be even more. So you can spend literally hours a day trying to keep up with all of this AI news and what it means for your career, your company, your department. Don't do that. All you got to do Join Us on Mondays we bring you the AI news that matters here on this very podcast as well as in our newsletter. But you can also just read our newsletter every single day. So each and every day we recap the podcast highlights, maybe something you missed, some additional content. But each and every day we do just bring you some of the top AI news stories in our newsletter.

Jordan Wilson [00:47:41]:
So don't waste time, just go to our website@youreverydayai.com so if this was helpful. If you're listening on the podcast, please leave us a rating. Please subscribe to the show. We would really appreciate that and I'd appreciate it if you join us tomorrow and every day for more everyday AI. Thanks y'all.