The Good, The Bad and The Ugly in AI this week 29/06/26
The Good, The Bad and The Ugly in AI this week 29/06/26
News from Around our schools
Celebrating Our AI Literacy Course Graduates
by Ayten Carvalho Pais, the AI Champion at Nobel Algarve
Congratulations to our 7 students who have successfully completed the AI Literacy Course, scoring 85% in the test and earned their certificate.
Alex, Max, Ivanna, Maria, Ricardo, Mels, Eason, Ruby, Isla, Kim, Eva-Rose, Nicandro, Makar, Martim, Amber, Mo Mo, Sophia, Yasmin, Emily, Eloise-Rose
Throughout the course, students explored the fundamentals of Artificial Intelligence, including how AI is used in everyday life, responsible and ethical AI practices, and the opportunities this technology presents for the future.
By completing the course, these students have developed valuable digital literacy skills and a deeper understanding of the technologies shaping our world.
We are proud of their achievement and look forward to seeing how they continue to apply their knowledge as confident and responsible users of AI.
This is a fantastic initiative Ayten. Your leadership at Nobel Algarve is wonderful to see. They are fortunate to have you doing this important work.
🌟 THE GOOD
🛍️ Pinterest launches AI shopping app
The Rundown: Pinterest has launched an experimental standalone app, Ask Pinterest, letting US users shop through a conversational chatbot instead of keyword search. It taps the company's Taste Graph, the internal map linking people to their interests and aesthetics.
The details:
• The app handles multi-step requests a normal search can't, such as planning a dinner party or furnishing a room over time, and remembers context across sessions.
• Pinterest also rolled out MCP and new AI advertising tools ahead of Cannes Lions.
• The company is keeping its data in-house rather than licensing it to rival AI shopping agents from Google, OpenAI and Meta.
Why it matters: Building a separate app lets Pinterest test ideas without breaking the main experience 500 million people already use. It's a measured way to trial conversational commerce while keeping its proprietary interest data behind its own walls.
🔊 Google launches smart Home speaker
The Rundown: Google opened pre-orders for its new $100 Home speaker, shipping 25 June. It delivers 360-degree audio in a compact, recycled-fabric body, comes in four colours, and pairs with a second unit or Google's TV Streamer for stereo.
The details:
• Gemini replaces the old Google Assistant as the built-in helper, handling multi-step commands like dimming the lights and setting a timer in one breath.
• It also remembers context for longer, so you stop repeating yourself and saying “Hey Google” constantly.
• The best features sit behind Google Home Premium at $10 or $20 a month, with a six-month trial included.
Why it matters: It's a capable, affordable device, but the subscription model is worth watching: after the trial, the speaker that talks naturally needs a paid plan to keep talking naturally — a pattern of locking core functionality behind recurring fees that's becoming common across consumer AI hardware.
🎨 Adobe puts an AI helper in everything
The Rundown: Adobe has added its Firefly AI assistant directly inside Premiere, Illustrator, InDesign and Frame.io, launching as a public beta. The assistant appears as a sidebar in each app to handle creative tasks.
The details:
• It takes on repetitive work: in Premiere it sorts clips into bins, batch-renames files and adds markers; in Illustrator it reorganises layers and checks for missing fonts.
• Firefly also gained brand-kit and storyboard generation.
• Adobe says ChatGPT, Claude and Copilot already work with Firefly, with Gemini and Slack coming soon.
Why it matters: For educators and students using the Adobe suite, embedded assistants could lower the admin burden of creative projects. The pitch increasingly mirrors Canva, suggesting Adobe would rather copy a rival than lose to it.
🧠 Perplexity launches a new memory system
The Rundown: Perplexity launched Brain, a memory system for its Computer agent. Instead of storing your name or preferences, Brain logs what the agent actually did: which sources held up, what users corrected, and what failed.
The details:
• Every task feeds a context graph. Overnight, Brain synthesises that graph into a personal LLM wiki and loads it into Computer's sandbox before the next task, so the agent skips work it already figured out.
• Perplexity claims 25% better correctness on repeat tasks, though those figures are first-party with no outside benchmark.
• The catch: Brain only helps Computer users already paying $200 a month, and Perplexity keeps all your data.
Why it matters: Task-level memory — learning from what worked rather than storing personal facts — is a genuinely interesting design direction for agents. But the unverified performance claim and the data-retention model deserve scrutiny before anyone treats the 25% figure as established.
🖼️ Getty's photos are coming to ChatGPT
The Rundown: Getty signed a multi-year deal letting OpenAI show its licensed photos inside ChatGPT search and discovery results, so a chatbot answer might now arrive with a properly licensed image attached.
The details:
• The move marks a sharp reversal for Getty, which banned AI-generated art in 2022 and later sued Stability AI for copyright violations.
• It struck a similar image deal with Perplexity back in October 2025.
• Investors cheered, sending Getty's stock up nearly 150% in pre-market trading.
Why it matters: Licensed imagery inside chatbots is a step towards properly compensating creators rather than scraping their work. One quieter detail keeps it out of the unreservedly good column, though: Getty won't say whether OpenAI can train on its images, even though its Perplexity deal flatly banned that.
🕶️ Meta launches $299 in-house AI smart glasses
The Rundown: Meta and EssilorLuxottica launched Meta Glasses starting at $299, the first pair to drop Ray-Ban and Oakley branding entirely. The display-less specs ship today across more than 10 countries in three frame styles.
The details:
• The new line undercuts Meta's $379 second-gen Ray-Ban glasses by about $80.
• It runs Muse Spark, the first AI model from Meta's Superintelligence Labs, which handles translation, scene recognition and everyday questions.
• Meta already owns most of the smart-glasses market, but Google, Samsung and Apple are all circling.
Why it matters: Cheaper AI wearables will accelerate adoption — and the safeguarding and privacy questions schools already face around always-on cameras. Pushing the entry price to $299 looks less like generosity and more like Meta building a moat before rivals arrive.
🤖 Anthropic launches Claude Tag for Slack
The Rundown: Anthropic launched Claude Tag, letting teams tag @Claude inside Slack channels to hand off work. The agent breaks each task into stages, runs them with approved tools, and reports back in the thread.
The details:
• Claude Tag runs on Opus 4.8 and replaces the older Claude in Slack app for Enterprise and Team customers.
• It builds memory per channel, works across days, and flags updates itself when admins enable ambient mode.
• Anthropic says 65% of its product team's code now comes from an internal version of the tool.
Why it matters: Embedding an agent in the place teams already collaborate lowers the barrier to delegation — relevant for school operations and admin teams weighing where AI actually saves time. As ever, the headline productivity figure is the vendor's own and worth treating as a claim rather than a finding.
🩺 China just landed two big medical AI wins
The Rundown: Two Chinese medical AI wins landed in one week. MicroPort MedBot's Toumai became the first remote surgical robot to earn the EU's CE mark, while Baichuan's M4 model topped OpenAI's HealthBench medical benchmark.
The details:
• The CE mark opens 30-plus European markets and follows the UK's first telesurgery using Toumai in March.
• Baichuan reports M4 beat GPT-5.5 by 15.9 points on HealthBench Hard with a 3.3% hallucination rate.
• HealthBench comes from OpenAI and uses rubrics from 262 doctors across 60 countries.
Why it matters: A Chinese model winning OpenAI's own medical test the same week a Chinese robot cleared Europe makes for tidy timing, and points to how quickly the centre of gravity in applied medical AI is shifting. The benchmark figures are Baichuan's own, so independent replication is the thing to watch for.
🏫 Norway restricts AI use in schools
The Rundown: Norway is largely barring generative AI for pupils aged 6 to 13 from late August, citing a risk that children skip foundational learning. Teens aged 14 to 16 may use it only under teacher supervision, and 17-plus students are encouraged to use it.
The details:
• The move follows declining national test scores and mirrors Norway's 2024 smartphone ban.
• The government also plans to fund more physical books in classrooms, reversing a decades-long shift towards tablets, and wants to bar under-16s from social media.
• Norway's earlier smartphone ban is reported to have cut bullying, lifted grades and reduced psychologist visits by around 60%, especially for girls.
Why it matters: This is the week's most significant story for education: a national system making an evidence-informed, age-graduated decision to protect foundational learning rather than reaching for blanket adoption or blanket bans. One honest caveat from the source — a school-hours rule does nothing about what children do at home — and the cited outcomes from the smartphone ban are worth verifying against the primary evaluation before quoting them.
⚠️ THE BAD
💸 OpenAI lost nearly $39 billion last year
The Rundown: Leaked audited documents show OpenAI's revenue more than tripled to $13.07 billion in 2025, yet expenses swallowed it whole. The company posted a $20.92 billion operating loss and a net loss of nearly $39 billion.
The details:
• Research and development alone hit $19.18 billion, including $10.59 billion paid to Microsoft.
• A one-time accounting charge tied to OpenAI's for-profit conversion inflated the headline loss; stripping it out drops the figure to roughly $8 billion.
• OpenAI tells investors it will turn profitable by 2030, but it just killed Sora and is cutting “side quests.” Of its 900 million weekly users, only about 50 million actually pay.
Why it matters: The economics underpinning the tools schools are being urged to adopt remain deeply unsettled. A roughly 1-in-18 paid-conversion rate and losses on this scale are a reminder to treat any vendor's long-term roadmap, and pricing, as provisional.
📉 Meta CTO says staff morale near all-time low
The Rundown: Meta CTO Andrew “Boz” Bosworth told staff on a 2 June internal call that morale ranks among the worst in 20 years, rivalling the 2018 Cambridge Analytica scandal, after months of layoffs and AI upheaval.
The details:
• May's layoffs cut about 8,000 workers, roughly 10% of staff.
• A March reshuffle swept 6,500 employees into a new Applied AI unit that many never asked to join.
• Bosworth now pledges bigger budgets for travel, events and snacks.
Why it matters: The human cost of the AI race is showing up inside the companies driving it, not just downstream. The detail that the layoffs landed weeks after Meta booked a $26.8 billion quarter is the part worth sitting with.
🤖 Americans use AI daily but don't trust it
The Rundown: A new Pew Research survey of 5,119 US adults found that just 16% expect AI to benefit society over the next 20 years, while 40% expect a negative impact.
The details:
• Around 67% doubt the government will regulate AI, 59% distrust companies to build it safely, and roughly two-thirds say development is moving too fast.
• Half of Americans still use chatbots anyway, led by ChatGPT at 44% and trailed by Claude at 6%.
Why it matters: Daily use alongside deep distrust is the defining public posture towards AI right now — and it's the gap educators have to work in. Note the figures are US-specific; UK and European attitudes differ, so this isn't directly transferable to our context.
📱 Apple to raise prices on chip shortage
The Rundown: Apple will raise prices across its products to offset surging memory and storage chip costs, CEO Tim Cook told the Wall Street Journal, calling the increases unavoidable but giving no timing, figures or affected models.
The details:
• An AI-fuelled scramble for data-centre parts has drained DRAM supply, as chipmakers route more high-bandwidth memory to AI servers.
• Cook said Apple may even tap its balance sheet to help expand capacity.
• Cook hands the reins to John Ternus on 1 September, as the foldable iPhone and iPhone 18 Pro arrive.
Why it matters: The AI build-out is now pushing up the price of ordinary consumer devices — a downstream cost that lands on school IT budgets and families alike. With no figures attached yet, the scale of the rises is still unknown.
🏦 Lloyds hires 300 AI staff, hints at cuts
The Rundown: Lloyds Banking Group is recruiting 300 tech experts to build and run agentic AI by September, joining a planned 1,000-strong AI team. The push comes weeks before CEO Charlie Nunn unveils a new multi-year strategy.
The details:
• The recruits will work on fraud detection, internal document search and a more personal app that lets customers ask plain-language money questions.
• They will deploy models like Anthropic's Claude and build on Google's Gemini.
• Generative AI added £50m to Lloyds last year, and the bank expects £100m this year.
Why it matters: A major UK employer is staffing up on AI while declining to rule out future job cuts — echoing Standard Chartered's 7,000 cuts last month. For careers educators, it's a concrete example of how AI is reshaping graduate and professional pathways in finance.
🎬 Amazon drops nearly finished Sam Altman film
The Rundown: Amazon MGM dropped Luca Guadagnino's nearly finished film “Artificial,” which stars Andrew Garfield as Altman and dramatises the chaotic 2023 days when OpenAI's board fired him and then rehired him.
The details:
• The decision came from Amazon studio chief Mike Hopkins, four months after Amazon committed $50 billion to OpenAI in a partnership expanding the lab's use of Amazon Web Services.
• Puck broke the news on Friday.
• Test screenings reportedly went well, but insiders say the film portrays Altman and Musk as its least likeable characters.
Why it matters: Shelving a finished film about your new $50 billion partner is a stark illustration of how commercial entanglement can shape what does and doesn't get made. The conflict-of-interest reading is the source's inference rather than a confirmed motive, but the timing is hard to ignore.
📱 Old iPhones get an unfixable security flaw
The Rundown: Researchers at Paradigm Shift published an exploit in the boot code of Apple's A12 and A13 chips that no software update can fix. It covers the iPhone XS through iPhone 11, plus some iPads and Apple Watches.
The details:
• The flaw lives in a USB-controller bug that lets an attacker write to off-limits memory and run code before iOS loads.
• A11 and A14-and-later chips handle that memory correctly, leaving A12 and A13 in a vulnerable middle.
• An attacker needs physical access, a cable and DFU mode, so this isn't remote, and it doesn't crack the Secure Enclave — passcodes and encrypted data stay safe.
Why it matters: Plenty of school and family hand-me-down devices fall in the affected range. The practical risk is limited because exploitation requires physical access, but the only true fix is a newer phone — a real cost for those least able to upgrade.
💀 THE UGLY
🩻 Midjourney just started a health company
The Rundown: Midjourney launched a health division and its first hardware product: a full-body ultrasound scanner that lowers you into water ringed with roughly 500,000 sensors and maps your insides in 60 seconds.
The details:
• The scanner runs on Butterfly Network's ultrasound-on-chip technology, which Midjourney licensed for $15 million upfront.
• The company will install ten of them in a San Francisco wellness spa opening late 2027.
• Midjourney says the scans rival MRI, but only about a dozen people have tried it and no peer-reviewed data backs that up.
Why it matters: An AI-image company making MRI-grade claims with no peer review, then launching as a spa rather than a clinic to dodge FDA rules, is exactly the kind of unproven medical marketing that erodes public trust. The “rivals MRI” claim is the company's own and currently unverifiable.
🩺 Americans now ask AI before asking a doctor
The Rundown: A West Health-Gallup survey found 1 in 4 US adults, about 66 million people, have asked an AI chatbot for physical or mental health information. Most use it to supplement care, not replace a doctor.
The details:
• Among adults earning under $24,000 a year, 32% turned to AI because they couldn't afford a doctor, compared with just 2% of those earning $180,000 or more.
• Only 4% strongly trust AI's health answers.
• 1 in 10 users say they got advice they considered dangerous, yet roughly 14 million skipped a doctor's visit anyway.
Why it matters: This is an equity story as much as a safety one: the people most likely to lean on AI for health advice are those who can least afford an alternative, and some are receiving advice they themselves judge dangerous. It's a sharp reminder that access gaps don't disappear with new technology — they migrate.
🤖 GM laid off 1,000 workers, then hired robots
The Rundown: GM laid off more than 1,000 workers at its Factory Zero EV plant in Detroit, then installed about 50 Fanuc cobots that bolt body panels onto vehicles alongside the workers who stayed.
The details:
• GM estimated the 2023 UAW contract added about $500 in labour costs per vehicle.
• UAW Local 22 filed grievances, and president James Cotton said members are “disgusted.”
• GM isn't alone: Toyota put Digit humanoids in its Ontario RAV4 plant and BMW is expanding its robot pilot to Germany.
Why it matters: The substitution of people with machines is happening in plain sight, and the framing — lay off the workers, then bring in the robots — is about as blunt as it gets. “Robots don't renegotiate every four years” captures exactly why this matters for how we prepare young people for the labour market.
📱 New TikTok users see mostly AI-made videos
The Rundown: A Kapwing study found 59% of the first 500 videos TikTok served a brand-new account were AI slop — low-quality, machine-made clips. That triples YouTube's rate of 21% in the same test.
The details:
• The numbers come from a cold-start feed, before TikTok learns a user's tastes.
• Across a wider 10,742-video sample, kids' content topped every category at 57%, while fitness, music and fashion stayed under 2%.
• In the #CartoonKids tag, 97 of 100 videos were AI-made.
Why it matters: The single most AI-saturated corner of the platform is the one aimed at children. For anyone thinking about young people's media diet, a default feed that is overwhelmingly synthetic — and most synthetic precisely where the youngest users are — is a serious concern. These are one firm's figures from a specific test, so treat the exact percentages as indicative.
📊 PENDING
🤖 Seres unveils its first humanoid robot Xiaosai
The Rundown: Chinese automaker Seres revealed Xiaosai, its first humanoid robot, built for visual recognition, autonomous greeting and voice interaction. Executive Kang Bo says several more embodied-intelligence robots will debut later this year.
The details:
• Xiaosai joins a factory that already runs over 3,000 industrial robots and 1,600 coordinated smart devices.
• Seres also signed a robotics deal with ByteDance unit Volcengine back in 2024.
• Nearly every Chinese automaker is chasing humanoids now: Xpeng's CEO just named himself robotics boss, BYD wants to sell robots through dealers, and Chery-backed Aimoga already retails one for about $42,260.
Why it matters: China's automakers are pivoting hard into humanoids, and the pace suggests embodied AI is moving from demo to product faster than expected. Whether Xiaosai amounts to more than a greeter remains to be seen.
🦊 Mozilla shares its plan to fix Firefox
The Rundown: Mozilla published a public Firefox roadmap alongside the Firefox 152 release, promising a Nova redesign, mobile VPN, customisable keyboard shortcuts, better PDF editing, and AI tools users can switch off entirely.
The details:
• The push answers a steady market-share slide, from 5.88% in May 2025 to 3.79% in May 2026 per Statcounter.
• Firefox 152 already added HDR video and experimental JPEG XL support.
• HDR playback arrives after users requested it for over six years on Mozilla's bug tracker.
Why it matters: Optional, switch-off AI is a notably user-respecting stance worth watching as a counterpoint to the always-on default elsewhere. But shipping a six-year-old request alongside an ambitious roadmap hints at why users drifted off in the first place.
🚀 NASA picks Relativity Space for 2028 Mars orbiter
The Rundown: NASA partnered with Relativity Space, run by former Google CEO Eric Schmidt, to fly a Mars orbiter called Aeolus in 2028. Relativity will build, launch and operate the spacecraft carrying NASA's instruments.
The details:
• Aeolus carries four instruments to deliver the first daily, global view of Martian winds, temperatures, dust and clouds.
• NASA says the data will help reduce risks for future crewed and uncrewed landings.
• Relativity will fly Aeolus on Terran R, a rocket that has yet to reach orbit.
Why it matters: An exciting mission resting on an unproven rocket and a 2028 timeline — hence Pending. If it launches on schedule, an unproven company could beat SpaceX to Mars.
🤖 Hyundai now owns Boston Dynamics
The Rundown: Hyundai Motor Group is buying SoftBank's remaining 9.65% stake in Boston Dynamics for $325 million, making the Waltham, Massachusetts robotics firm a wholly owned subsidiary.
The details:
• The sale follows a put option SoftBank kept when Hyundai took control in 2021.
• Hyundai paid about $880 million for 80% back then, valuing the company near $1.1 billion.
• Hyundai plans to put a production Atlas humanoid to work at its Georgia EV plant by 2028.
Why it matters: Owning both the factories and the robot maker means Hyundai doesn't have to hunt for its first customer — a vertically integrated route into industrial humanoids that others will likely follow.
📉 SpaceX shares drop 6.5% after IPO surge
The Rundown: SpaceX shares fell 6.5% on Thursday to $178.50, knocking roughly $150 billion off a market cap that had topped $2.5 trillion. The stock still traded about 30% above its $135 IPO price.
The details:
• SpaceX raised $75 billion the week before in the largest IPO ever, briefly cracking the world's five most valuable companies.
• Retail investors who piled in for three sessions barely bought on Thursday, netting just $9 million.
• This week it agreed to buy Cursor maker Anysphere for $60 billion and lined up a bond sale of at least $20 billion.
Why it matters: Investors are weighing SpaceX's costly AI pivot, and the early volatility says the market hasn't settled on a value. One to watch rather than read too much into yet.
🤖 Google builds its own AI chip ecosystem
The Rundown: Google is copying Nvidia's playbook, pairing its custom TPU chips with cloud software and developer tools to lock in an ecosystem. In April it launched eighth-gen TPUs split into separate training and inference models.
The details:
• Nvidia built its dominance by pairing GPUs with CUDA software and developer lock-in, and still holds over 80% of the roughly $400 billion AI chip market.
• Google is targeting that moat with cheaper, cloud-optimised silicon.
• Meta and Anthropic have both signed major TPU deals, with Anthropic alone reportedly committing up to a million chips.
Why it matters: Competition at the silicon layer could ease the Nvidia bottleneck that shapes AI costs everywhere downstream. When rivals start becoming customers, it's an early sign the strategy has traction.
🤝 OpenAI hires Gemini co-lead Noam Shazeer
The Rundown: Noam Shazeer announced on 18 June that he is leaving Google to join OpenAI. He co-led Google's Gemini models as a VP of engineering and co-authored the 2017 transformer paper behind most modern chatbots.
The details:
• Google reportedly paid $2.7 billion in 2024 to bring Shazeer back from Character.AI, the startup he founded after Google passed on his early chatbot pitch.
• He walked out again under two years later.
• His exit lands as OpenAI heads towards a reported $1 trillion IPO.
Why it matters: The churn of senior talent between the frontier labs continues to reshape who builds what. Google's $2.7 billion bet shows the catch with acqui-hires: you can buy the talent, but you can't make it stay.
🗣️ Tencent tests an AI assistant inside WeChat
The Rundown: Tencent started testing Xiaowei, an AI assistant baked into WeChat, with a small group of users. It takes text or voice commands to send messages, place calls and run the app's mini-programs.
The details:
• Tencent is chasing rivals like Alibaba, DeepSeek and Zhipu.
• WeChat's 1.4 billion users give it a built-in audience.
• Xiaowei mainly runs on Tencent's own WeLM model, leaning on DeepSeek for some queries.
Why it matters: Embedding an assistant inside the world's most-used super-app could be a major distribution moment for conversational AI. The twist: Tencent leans on DeepSeek, one of the very rivals it's racing to catch, to power part of it.
📉 SpaceX shares fall from $225 to below $150
The Rundown: SpaceX stock briefly fell to $149 on Tuesday, slipping under its $150 opening price for the first time since the 12 June IPO. Shares then recovered, climbing 1.3% to around $156.
The details:
• The slide marks a 34% drop from the 16 June peak of $225.64.
• A broad chip selloff dragged markets lower on Tuesday, with Micron sinking 9.2% ahead of its earnings report.
• Morningstar pegs fair value near $62.
Why it matters: The second SpaceX share story this week underlines how unsettled the post-IPO valuation is. With an analyst fair-value estimate far below the trading price, a good deal of the remaining gains may yet prove shaky.
🎬 Google invests $75M in film studio A24
The Rundown: Google DeepMind is investing roughly $75 million in indie studio A24, the company behind Everything Everywhere All At Once. The deal marks Alphabet's first equity stake in a film studio.
The details:
• The partnership pairs DeepMind researchers with A24's tech team to build filmmaking tools, starting with AI-generated storyboards.
• The agreement stays non-exclusive and locks Google out of A24's film library and content data.
• DeepMind already runs Veo, one of the strongest AI video generators.
Why it matters: DeepMind hardly needs A24 for the technology; what it really buys is credibility with artists who mostly distrust the tech. Whether that goodwill holds is the open question.
🔌 SpaceX rents its AI computers to rivals
The Rundown: SpaceX signed a deal letting Reflection AI rent Nvidia GB300 chips at its Colossus 2 data centre, paying $150 million monthly from July through 2029 and reaching up to $6.3 billion if the full term runs.
The details:
• Colossus started as the training engine for Musk's Grok models, but SpaceX now rents spare capacity to outside labs.
• Anthropic reportedly pays $1.25 billion monthly and Google $920 million monthly for similar access.
• Reflection still hasn't shipped a public model.
Why it matters: The scale of these compute commitments — billions a month — shows how much of the AI economy is now infrastructure rental. That a lab with no public model can commit billions in compute is a sign of how speculative the spending has become.
🖥️ China beats US with world's fastest supercomputer
The Rundown: China's LineShine debuted at No. 1 on the June 2026 Top500 list, hitting 2.198 exaflops and beating El Capitan's 1.809. It hands China the top spot for the first time since 2017.
The details:
• LineShine skips GPUs entirely, running on roughly 14 million cores from a custom 304-core Chinese processor.
• That design let it lead the list despite US export controls limiting China's access to advanced AI chips.
• On the AI-style test it placed fourth, behind El Capitan.
Why it matters: A reminder that export controls reshape rather than simply halt progress. The crown says more about raw computing power than AI muscle, but the geopolitical signal is real.
🔮 Meta is building a prediction market app
The Rundown: The New York Times reports Zuckerberg has told a small Meta team to build a Polymarket-style app, codenamed Arena. It would run separately from Facebook and Instagram, but Meta would funnel its own users towards it.
The details:
• Arena would skip real money at first, running on a video-game-style points system, though Meta hasn't ruled out cash bets later.
• It chases Polymarket and Kalshi, which now clear tens of billions in trading volume.
• Meta tried this once before with Forecast, launched in 2020 and killed in 2022.
Why it matters: A platform of Meta's reach steering users towards prediction markets raises obvious questions about gambling-adjacent design and young users. A points-only app earns nothing, so the real bet is that Meta's billions of users eventually want to wager cash.
📺 Instagram tests episodic shows and live TV
The Rundown: Instagram is exploring longer-form videos, episodic series and Live TV for its television app, pushing past short Reels to compete with Netflix and Amazon Prime Video for living-room attention.
The details:
• The expansion lands as Instagram for TV reaches Samsung sets across the US, joining Amazon Fire TV and Google TV.
• New features add interest-based channels, phone-to-TV casting, horizontal video and Stories on the big screen.
• Meta is already recruiting creators to make episodic content.
Why it matters: The battle for screen time is moving back to the living room, with social platforms chasing the microdrama format TikTok moved into first. Whether a 30-second Reels star can hold a couch audience to minute 20 is the open question.
🎮 GTA VI preorders start today, $80 base price
The Rundown: Rockstar priced GTA VI at $80, with a $100 Ultimate Edition, and opened preorders on 25 June before its 19 November launch. That matches Nintendo's Mario Kart World, not the $100 some analysts had floated.
The details:
• The game stays exclusive to PS5 and Xbox Series X|S, with no PC or Switch 2 version at launch.
• Every copy is digital, so even physical boxes just hold a download code instead of a disc.
• Rockstar says the $100 edition threads exclusive “action” across Jason and Lucia's whole story.
Why it matters: Not strictly an AI story, but a bellwether for premium pricing and digital-only distribution that shapes the games landscape young people inhabit. The $20 “action” tier reads less like bonus cosmetics and more like chunks of the actual game behind a paywall.