Tech Brew Ride Home - Now Meta Has Frozen AI Hiring

Episode Date: August 21, 2025

What is going on at Meta AI story number 77. A deeper look at that new chip that Google unveiled with the Pixel phones yesterday and what that could suggest for Google’s hardware strategy. And an interesting startup that is making AI smartglasses happen now. Links: Meta Freezes AI Hiring After Blockbuster Spending Spree (WSJ) Suddenly, Silicon Valley Is Lowering AI Expectations (NYMag) Google did more than just switch to TSMC for Tensor G5 — This is what’s new (9to5Google) DeepSeek Touts Model That Outdoes Flagship in Agentic AI Step (Bloomberg) Sony is raising PS5 prices, starting tomorrow (The Verge) Harvard dropouts to launch ‘always on’ AI smart glasses that listen and record every conversation (TechCrunch) Learn more about your ad choices. Visit megaphone.fm/adchoices

It's peak pollination season, and my business is scaling fast. To keep the nectar flowing, I need a phone plan with top priority data speed. That's why I chose GoogleFi Wireless. My connections stay strong even when the hive is buzzing. Plus, unlimited plans start at $35 a month. Now, that's a deal that doesn't stay. Explore GoogleFi Wireless plans today. Plus taxes and government fees.

GoogleFi Wireless is not subject to data traffic deprioritization during times of high network usage. Welcome to the Tech Brew Ride Home for Thursday, August 21st, 2025. I'm Brian McCullough today. What is going on at Meta-I story number 77? A deeper look at that new chip that Google unveiled with the pixel phones yesterday and what that could suggest for Google's hardware strategy going forward and an interesting startup that is making AI smart glasses happen now. Here's what you missed today in the world of tech. Study and play. Come together on a Windows 11 PC. And for a limited time, college students get the best of both worlds. Get the unreal college deal, everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 premium and a year of Xbox GamePass Ultimate with a custom color Xbox wireless controller. Learn more at Windows.com slash student offer. While supplies last, ends June 30th, terms at AKA.ms.

slash college PC. Well, I did not have this on my bingo card, even though we have been speculating about this as it's been happening. I know at some point I said that if Mark Zuckerberg signals he's slowing spending on AI or just evidences, some sort of cooling enthusiasm for AI generally, well, look out for people's fears of bubbles popping. And, well, for like the third day in a row now, Zuck is delivering headlines like this. The journal says that meta froze hiring in its

AI division last week, with sources saying new hires now need Alexander Wang's permission. Quote, the hiring freeze, which went into effect last week and coincides with a broader restructuring of the AI group, also prohibits current employees from moving across teams inside the division. The duration of the freeze wasn't communicated internally. There might be some exceptions to the block on external hires, but they would need permission from Meta's chief AI officer Alexander Wang, the people said. a meta spokesperson confirmed the freeze, characterizing it as, quote, basic organizational planning, creating a solid structure for our new superintelligence efforts after bringing people on board

and undertaking yearly budgeting and planning exercises. At least three members of the previous AGI Foundation's team said internally they were leaving the company around the time of META's latest vesting date, August 15th, according to internal posts viewed by the Wall Street Journal. As of mid-August, META had successfully hired more than 20 researchers and engineers from Open AI for the effort, at least 13 from Google, three from Apple, three from XAI, and two from Anthropic for a total of 50-plus new employees. Mounting concern from investors over the costs of the tech giants' AI buildout has played a role in this week's sell-off of technology stocks.

In an August 18th research note, analysts at Morgan Stanley warned that the fast-rising stock-based compensation offered by Meta and Google to lure AI talent could threaten their ability to return capital to shareholders via buybacks. Lavish spending on talent, the analysts wrote, quote, has the potential to drive AI breakthroughs with massive value creation or could dilute shareholder value without any clear innovation gains, end quote. Now, I want to point out that Andy Stone met his head of communications took to X and accused the Wall Street Journal of not printing their full statement, which apparently renders the

report sort of a nothing burger and totally not damage control as Eurohedge, snarkily pointed out, quoting Stone, mundane, because all that's happening is a temporary pause on some hiring while we do planning and forecasting, which we do regularly and a structure for our new superintelligence efforts after we brought a number of new people on board, end quote. But, quoting New York Magazine, it is not just Zuck tamping down on expectations for AI all of the sudden. You also have the general disappointment of the GPT-5 rollout and Sam Altman's recent talk about over exuberance and bubbles that we've already discussed. There was also some interesting quotes from another interesting person, quoting New York Mag. In another notable shift, former Google

CEO Eric Schmidt, who has in recent years become an influential and provocatively severe figure in the AI world, co-authored a piece for the New York Times with tech analyst Selena Zhu, arguing that Silicon Valley quote, needs to stop obsessing over superhuman AI. Just a few months ago, In another op-ed, Schmidt argued that, quote, even without a consensus about a precise definition, the contours of an AGI future are beginning to take shape, suggesting that, quote, artificial general intelligence could usher in a new renaissance. This week, he and Zhu urged their peers to rethink such messaging and suggested that narratives like this actually give them pause. As an insider effort to regulate the frequently wild messaging coming from the AI industry,

this is surprising rhetoric. It is consistent with and actually cites critical, work arguing that while recent developments in AI represent a major technology with great disruptive potential, LLMs might not be on the cusp of rendering the entire economy obsolete and or wiping out or subjugating humanity once it has established itself as a superior species. The argument that stories about AGI are alienating, distracting, and disempowering, no objection here, is likewise sort of wild to hear coming from Schmidt, who just this year warned that AI models are beginning recursive self-improvement and are already, quote, learning how to plan, and they don't have to listen to us anymore, end quote. But political scientists

and AI theorist Henry Farrell makes a case for the deeper significance of such a pivot from Schmidt, who last year co-authored a book with Henry Kissinger about AI and who Farrell has argued could be, quote, the most influential American foreign policy thinker of the early 21st century. In short, Schmidt was instrumental in forming the mind meld between Silicon Valley and national security policymakers, which held that self-improving AI and AGI represented a fast-approaching inflection point in a race between the U.S. and China, the winner of which would gain a technological advantage that would secure its long-term dominance. But he writes, quote, if the AGI bet is a bad one, then much of the rationale for this consensus falls apart. And that is the conclusion that

Eric Schmidt seems to be coming to. The other argument in Schmidt and Zhu's op-ed, that in China, where scientists and policymakers aren't as AGI-pilled as their American counterparts, citizens are far more approving of AI, which they understand as a useful tool rather than an abstract threat, which makes clear that this is intended as a narrative departure. Many of the purported benefits of AGI in science, education, health care, and the like, can already be achieved with the careful refinement and use of powerful existing models, they write. The belief in an AGI or superintelligence tipping point flies in the face of the history of technology in which progress and diffusion have been incremental, end quote.

Eric Schmidt and Sam Altman have distinct goals and motives for weighing in like this, of course, and in isolation, each statement could be understood as a bit of cautious hedging. But taken together and combined with Altman's recent assertions that AGI is, quote, not a super useful term, their not-so-sodeled repositioning could suggest a bit of a vibe shift coming among the tech elite, or at least a heightened awareness of how they sound to everyone without a financial stake in the AI boom. AI is still a huge deal, they're suggesting, but after years of issuing one sort of warning, they're now voicing another. Let's not get ahead of ourselves, end quote.

But wait, there's one more because Mustafa Suleiman has a blog post out this morning, essentially saying, forget about AGI, but get ready for what he calls seemingly conscious AI. Suleiman says, seemingly conscious AI, which would be convincing but non-sent AI systems, could emerge at any point now. Risking delusions, dependency, and rights debates, he urges that we agree now on a set of norms, like avoid implying consciousness generally, and developers should build AI for people, not for AI to be a digital person. So again, this is another pulling the goalposts back a bit post. As Dar Obisandro said on Blue Sky, quote, looks like there was a pump the brakes on the AI hype group chat this weekend.

Microsoft's AI chief Mustafa Soleiman warns against building AI that seems conscious, saying it fuels, quote, AI psychosis. He argues the industry should stick to assistance, not AGI. Altman, Schmidt, and now Suleiman, end quote. I did want to come back to something from yesterday's pixel event that we didn't go into very much detail about, and that is the TensorFlow G5, that new three nanometer chip manufactured by TSMC, which Google says has significant gains in, performance and runs the newest Gemini Nano model locally. Quoting 9 to 5 Google.

Google says to expect improvements in responsiveness when browsing the web, launching apps, OS rendering, AI experiences, and other workloads because of the new chips. There's one big performance core, five mid-performance cores, and two efficiency cores. Google has also upgraded hardware and software thermal controls to let the chip operate at higher frequencies without throttling. TensorFlow G5 has a high-speed memory interface, thanks to LPDR 5X, higher memory bandwidth, and UFS 4.0 faster flash storage, end quote. Apparently, the TensorFlow G5 isn't just another generational upgrade.

It's shaping up to be Google's most meaningful leap in their custom silicon yet. Built on TSM's Leaning Edge 3-Nameter node, as mentioned, rather than Samsung. The G5 benefits from both tighter power efficiency and raw performance gains, a strategic manufacturing shift that signals deeper ambitions for Pixel's hardware future. CPU performance sees a solid boost with Google citing a 34% average gain. This is thanks to that reconfigured 1 plus 5 plus 2 core layout, 1 performance core 5 mid-tier cores to efficiency cores. Add to that, upgraded thermal controls, graphite for the base model and vapor chamber cooling for pro variants, the G5 can sustain higher clock speeds without throttling. But AI capabilities are where

Google, of course, wants you to know the Tensor G5 truly stands out. Its TPU is up to 60% more powerful, enabling the latest Gemini Nano model to run 2.6x faster and twice as efficiently on tasks like pixel screenshots and recorder compared to the G4. Even more impressive, the model now supports a 32,000 token context window over twice the window of last year, unlocking richer, long-form generative tasks like understanding a month's worth of emails or maybe 100 screenshots in one go. Other new AI features this allows include how Magic Q proactively surfaces helpful info. Voice translate works in real time. Call notes automates key actions and the new personal journal thing, all powered locally without sending data to the cloud. The G5

also brings advancements in computational photography, its redesign image signal processor delivers motion de blur, 10-bit HDR video by default, 1080p, and 4K30, and more. nuanced real tone rendering across different skin tones. In essence, with the TensorFlow G5, Google is leaning fully into tighter hardware-software synergies almost the way Apple does, if Google stands won't hate me for pointing that out. Rather than chasing benchmark dominance, the focus this time seems to be on delivering seamless, AI-driven experiences, more responsive, smarter to give the feel of next-gen, as opposed to just calling it next-gen. I don't think we mention this, but Deepseek recently unveiled V3.1 of their AI model, but they did it very quietly.

They say the new version surpasses R1 on key benchmarks and is customized to work with the next-gen Chinese-made AI chips. But this leads to the growing question, where is the R2 model? Quoting Bloomberg. The V3-1 returns answers to queries much faster and marks the startup's first step toward creating an AI agent. Deepseek said in a Wii chat post Thursday. The company first outlined the V3.1 earlier this week, but the platform only just made it to the Hugging Face portal. The version has been customized to work with Next Generation Chinese-made AI chips, the startup said in a separate message. Industry Watchers await the release of the successor to R1, though, which had been expected to

emerge earlier this year. Local media have attributed the delay to founder Liang Wen Feng's determination to get it right, even as he continues to run. his lucrative high-flyer asset management outfit. Others have speculated about various glitches in training or development. At the same time, rivals from Alibaba to Tencent have kept up a frenetic pace of AI model development and updates. Alibaba's Quen models have, in particular, gained a popular following, end quote. Ready to soundtrack your summer?

With Red Bull Summer All Day Play, you choose a playlist that fits your summer vibe the best. Are you a festival fanatic? A deep end DJ? a road dog or a trail mixer. Just add a song to your chosen playlist and put your summer on track. Red Bull Summer All Day Play. Red Bull gives you wings. Visit redbull.com slash bright summer ahead to learn more.

See you this summer. Sony announced yesterday it is raising PlayStation 5 prices by $50 across the board for all models in the U.S. due to what it calls a, quote, challenging economic environment. Quoting the verge. The changes will go into effect on 30. Thursday, which is today, by the way, and the new prices are as follows. The PlayStation 5 5 comes in at $549.99. The PlayStation 5 Digital Edition is $499.99. The PlayStation 5 Pro is 74999. Sony says that the retail prices for PS5 accessories, quote, remain unchanged. In April,

Sony raised the price of PS5 hardware in the UK, Europe, Australia, and New Zealand, and in May, the company said it was considering price hikes to cover the Trump administration's tariffs. has also raised the price of Xbox hardware and accessories, including bumping up the Xbox Series X digital edition to 549-99. Ahead of the Switch 2's launch, Nintendo hiked the cost of accessories for the console. And earlier this month, it raised the price of the original switch due to the impact of tariffs, end quote. Finally, today, I keep reading stories about people hacking Rayban meta-glasses to do cool or

perhaps creepy things. And I've not done one of those stories yet, so let this stand in for that, because this is something else entirely. Halo is a startup founded by former Harvard students who hacked a facial recognition app for Rayban meta glasses previously, and these new glasses will introduce what they're calling Always On AI for $249. Quoting TechCrunch. Two former Harvard students are launching a pair of Always On AI powered smart glasses that listen to, record, and transcribe every conversation and then display relevant information to the wearer in real time. Our goal is to make glasses that make you super intelligent the moment you put them on, said

Anfou Nguyen, co-founder of Halo, a startup that's developing the technology. Or as his co-founder, Kane Artifio put it, the glasses give you infinite memory. The AI listens to every conversation you have and uses that knowledge to tell you what to say, kind of like IRL Clule. Artifio told TechRens referring to the startup that claims to help users cheat on everything from job interviews to school exams. If somebody says a complex word or ask you a question like, what's 37 to the third power or something like that, then it'll pop up on the glasses. Artifio added. Artifio and Nguyen have raised $1 million to develop the glasses led by Pillar VC with support from Somac Capital, Village Global, and Morningside Ventures. The glasses will be

priced at $249 and will be available for pre-order starting Wednesday. Artifio called the glasses the first real step toward vibe thinking. The two Ivy League dropouts who have since moved into their own version of the hacker hostel in the San Francisco Bay area recently caused a stir after developing a facial recognition app for Meta's smart rayban glasses to prove that the tech could be used to docks people. As a potential early competitor to Meta's smart glasses, Artifio said Meta, given its history of security and privacy scandals, had to rein in its product in ways that Halo can ultimately capitalize on. Meta doesn't have a great reputation for caring about user privacy and for them to release something that's always

there with you, which obviously brings a ton of utility, is just a huge reputational risk for them that they probably won't take before a startup does it at scale first. Nguyen added. And while Nguyen has a point, users may not yet have a good reason to trust the technology of a couple of college-age students purporting to send people out into the world with covert recording equipment, while META's glasses have an indicator light when their cameras and microphones are watching and listening as a mechanism to warn others that they are being recorded. Artifio says that the Halo Glasses dubbed Halo X do not have an external indicator to warn people of their customers recording. For the hardware we're making, we want it to be discreet like normal glasses, said Artifio,

who added that the glasses record every word, transcribe it, and then delete the audio file. Privacy advocates are warning about the normalization of covert recording devices in public. Small and discrete recording devices are not new. Eva Galprin, the director of cybersecurity at the Electronic Frontier Foundation, told TechCrunch, in some ways this sounds like a variation on the microphone spy pen, said Galprin. But I think that normalizing the use of an always-on recording device, which in many circumstances would require the user to get the consent of everyone within recording distance, eats away at the expectation of privacy we have for our conversations

in all kinds of spaces. Artifio says Halo relies on soniaks for audio transcription, which claims to never store recordings. Nguyen claimed when the finished product is released to customers, it will be end-to-end encrypted, but provided no evidence of how this would work. He also noted that Halo is aiming to get SOC2 compliance, which means it has been independently audited and demonstrates adequate protection of customer data. A date for the completed SOC2 compliance was not provided. Still, the two students are not new to privacy-invasive controversial projects.

While still at Harvard last year, Artifio and Nguyen developed i.X-ray, a demo project that added facial recognition capabilities to the meta-rayband smart glasses, demonstrating how easily the tech could be bolted onto a device not meant to identify people. For now, HaloX glasses only have a display and a microphone but no camera, although the two are exploring the possibility of adding it to a future model. Users still need to have their smartphones handy to help power the glasses and get real-time info prompts and answers to questions per Nguyen. The glasses, which are manufactured by another company that the startup didn't name,

are tethered to an accompanying app on the owner's phone where the glasses essentially outsource the computing since they don't have enough power to do it on the device itself. Under the hood, the smart glasses use Google's Gemini and Perplexity as its chatbot engine, according to the two co-founders. Gemini is better for math and reasoning, whereas they use Perplexity to scrape the Internet, they said, end quote. Nothing more for you today. Talk to you tomorrow.

Ambition comes in all shapes and sizes. At First Citizens Bank, we roll with your goal. because we're built for what you're building. Fit for your ambition for citizens' backing.