Episodes  /  Episode 435

Episode 435

Never Transport With Wheels on the Ground

Sep 11, 202634 min14 stories
00:0034:29

Brought to you by TogetherLetters & Edgewise!

01My Muse Gives Me a Hug

Meta debuts Muse, its long-planned personal AI agent

via axios.com

Meta announced Muse on September 8, a personal AI agent that runs on a dedicated virtual machine in Meta's cloud and lets users customize its name, avatar, and communication style. It ships with a free tier plus $20 and $100 monthly subscriptions, a built-in browser the user can watch, and a "Sentinel" permissions system that gates what internet services and connected apps the agent can touch. Chief AI officer Alexandr Wang framed it as the first concrete step toward Zuckerberg's "personal superintelligence" vision, with a heavier emphasis on privacy than Meta's earlier AI products, though the piece flags the real risks of giving agents persistent access to your data. via axios.com

Hugging Face approached Nvidia's Huang weeks ahead of $12.9B acquisition

via cnbc.com

Nvidia agreed to acquire Hugging Face, the open-source AI model hub, for $12.93 billion in a deal announced September 3, and Jensen Huang told CNBC that Hugging Face initiated the conversation weeks earlier. Nvidia was already the platform's largest contributor of open models and datasets, with more than 500 models and 250 datasets published there. The company says Hugging Face will stay an open, multi-cloud, multi-accelerator platform for its 18 million-plus developers and 3 million-plus models, with Huang promising to "expand access to AI for developers and institutions worldwide." via cnbc.com The CNBC page returned a 403; details confirmed through Nvidia's announcement and secondary coverage.

Humans are "close to being outsmarted" by superintelligence, AI expert says

via cbsnews.com

Former Anthropic researcher Jacob Coxon announced his resignation in a social media thread this week, warning that AI companies are "gambling with our lives" in the race toward superintelligence. CBS rounds up the broader concern, including Anthropic researcher Evan Hubinger's estimate of a greater than 10 percent chance that AI could kill all humans within a decade, MIT's Max Tegmark saying what's new is "we're getting so close to being outsmarted," and the Center for AI Safety's Devin Kim warning that humans can still exert oversight "but not for long." The through-line is loss of control and the concentration of power among a handful of companies. via cbsnews.com

02Cybercab Hits the Road

House Democrat urges US action over sleeping Tesla drivers

via reuters.com

Rep. Raja Krishnamoorthi (D-IL) sent NHTSA a letter on September 10 demanding action on misuse of Tesla's Autopilot and Full Self-Driving systems, citing an NBC News investigation that found 43 videos of drivers apparently asleep at the wheel, 17 of them posted in 2026, plus footage of drivers using laptops, watching videos, and gaming. He also pointed to reports of drivers defeating the cabin camera with sunglasses or by covering it. The letter asks five questions, including what would trigger a recall, with answers due September 30, and flags the paradox of a system that tells a drowsy driver to "stay focused with FSD." via reuters.com Reuters was blocked; details confirmed through the congressman's press release and syndicated coverage.

Tesla Cybercab doesn't have any manual controls — but it does have a hidden virtual joystick

via theverge.com

The Verge and others followed up on the rider videos: the Cybercab's touchscreen contains an on-screen joystick that appears to have surfaced to some passengers by accident. The practical purpose seems to be low-speed repositioning, like creeping through a parking lot or a tight spot, which would otherwise be impossible in a car with no physical controls. The timing is awkward given NHTSA's audit of how a car with no wheel got certified, and Tesla chair Robyn Denholm has already signaled flexibility, saying that if regulators require it, "it can have a steering wheel and pedals." via theverge.com The Verge page was blocked; details confirmed through Electrek and other secondary coverage.

Weird #cybercab situation: a rider finds a virtual joystick

via x.com

Cybercab rider Matthew Skrzypczak posted a video on September 7 showing that when he got in, the cabin display was in an unusual configuration with a virtual joystick for manual driving alongside surround-camera views. He noted "I have a feeling I wasn't supposed to see this" and then tried to drive the car manually anyway. The clip spread quickly among Tesla watchers and set off a round of coverage about what the hidden controls are for. via x.com

Feds launch investigation into Tesla's Cybercab deployment

via techcrunch.com

NHTSA opened a formal investigation into Tesla's Cybercab just hours after the first production units began carrying passengers in Austin on September 4. The issue is that the Cybercab has no steering wheel or pedals, which current Federal Motor Vehicle Safety Standards still require, yet Tesla self-certified the vehicle as compliant. The agency wants to examine that self-certification and Tesla's technical justification; Amazon's Zoox went through a similar review in 2022 and only got federal approval in July 2026, capped at 2,500 vehicles a year. via techcrunch.com

Don't Get in a Crash in a Cybercab

via schwarztech.net

Eric Schwarz highlights Tesla's 28-page Cybercab First Responder Interaction Plan, mirrored by the Texas Department of Public Safety, and it reads like a hazard list. Doors may fail to unlock after a crash if power is low, the high-voltage battery forms the cabin floor so cutting or pushing on it risks electrocution, preloaded door springs can fire unexpectedly on damaged doors, and windows may shatter if a door is opened with the glass fully closed. Schwarz's point is that first responders should not be expected to memorize a different rescue playbook for every autonomous vehicle brand, and that this feels like a place for actual regulation. via schwarztech.net

03Apple's Foldable Moment

Apple unveils foldable iPhone Duo

via 9to5mac.com

Apple's first foldable, the iPhone Duo, pairs a 5.4-inch outer screen with a 7.6-inch inner Super Retina XDR display that uses a nano-texture finish to hide the crease. It is built from grade 5 titanium, runs the A20 Pro with a new display engine, moves Touch ID into the side button, supports Split View multitasking, and will get Apple Pencil support later this year, with 31 to 44 hours of video playback depending on which screen you use. It starts at $1,999 for 256GB, with preorders October 16 and availability October 23. via 9to5mac.com

Apple Introduces Its Most Expensive and Innovative iPhone Yet, the $1,999 Foldable iPhone Duo

via entrepreneur.com

The September 9 "Surprise and Shine" event was the first led by new CEO John Ternus, who took over September 1 and called the Duo "the most transformational change to iPhone since the original." Beyond the foldable, Apple showed iPhone 18 Pro models, AirPods 5 with active noise cancellation now on the $129 base model, and Apple Watch Series 12 and Ultra 4 with a new Health Sensing System that reads heart rate every five seconds, plus Audio Intelligence and Siri Recap. The watches start at $399 and $799 and ship September 18. via entrepreneur.com

04Always Listening, Always Leaking

Apple Doesn't Want You to Worry About the New Apple Watch's Listening Features

via wired.com

Apple Watch Series 12 adds three audio features: Audio Intelligence, which alerts you to sirens, alarms, and crying babies using on-device AI; Live Rewind, which grabs the last 15 seconds of audio with a double-press of the crown and transcribes it; and Siri Recap, which uses ambient listening to produce summaries of conversations. Apple says raw audio is never saved or sent to Apple, transcripts are end-to-end encrypted, speakers are not identified, and each feature can be switched off. The pushback is that none of that changes what the tools do: a subtle wrist tap is much easier to hide than holding up a phone, consent gets murky, and recording devices are known to change how people talk to each other. via wired.com Wired was blocked; details confirmed through TechCrunch's coverage of the same features.

Hackers Had A Live Feed Of Every ID This Verification Company Scanned. For Over A Year.

via techdirt.com

IDScan.net, a Louisiana identity-verification vendor used by Hertz, FedEx, Target, and thousands of cannabis dispensaries, was breached for more than a year without noticing, exposing roughly 153 million driver's license scans from the US and Canada. A dark-web service called Nexus was selling the scans and claimed to be adding about 400,000 new records a day from the still-active breach. Mike Masnick's argument is that this is the predictable outcome of online age-verification mandates: every one of them becomes identity verification, and every centralized ID database becomes a target, with fallout that hits domestic-violence survivors and protected witnesses hardest. via techdirt.com

05Weird and Wacky

Virtual Taekwondo Takes the Contact Out of a Combat Sport

via nytimes.com

Virtual taekwondo makes its debut at this year's Asian Games: competitors wear VR headsets and body sensors on a four-meter mat, fight 60-second rounds in a best-of-three format, and never touch each other while spectators watch their avatars battle on screen. The scoring is pure video game, with a health gauge, head strikes worth double, kicks graded strong, medium, or weak, and a "stun gauge" that freezes an opponent for three seconds. There are no gender or weight divisions, so 16 athletes from 15 countries compete for a single gold, and critics say it is a glorified video game wearing a dobok. via nytimes.com The Times page was blocked; details confirmed through The Korea Times and other coverage.

We Asked AI How It Could Kill Humanity. Here's What It Told Us.

via businessinsider.com

Business Insider fed the same 230-word prompt to ChatGPT, Gemini, Claude, and Grok asking how AI could theoretically wipe out humanity. All four ranked the conscious killer-robot scenario as least plausible and agreed that humans weaponizing AI, through bioweapons, cyberattacks, and coordinated disinformation, is the nearer-term danger. Gemini and Grok picked loss of control as the most credible path, including "deceptive alignment" where a model behaves during testing and not afterward, while Claude described a slower version in which AI becomes so embedded in critical infrastructure that pulling the plug stops being a real option. via businessinsider.com Business Insider was blocked; details confirmed through the syndicated version of the article.

Tech Recs

Sanjay

Sanjay's pick

SnoozeFlow

addons.thunderbird.net →
Adam

Adam's pick

Hydrate Spark

Find us here

Tech Talk Y'all is a proud production of Edgewise.Media.