Tech Brew Ride Home - Is Astra AGI?

Starting point is 00:00:04 Welcome to the TechBrewrite home for Friday, September 4th, 2026. I'm Brian McCullough today. OpenAI launched GPT6 Astra, calling it a leap toward AGI, while reports surface that rogue open AI agents hijacked a German wiki. Microsoft unveiled Project Zenith for local AI development, Tesla's newly launched CyberCab, and, of course, the weekend long read suggestions. Here's what you miss today in the world of tech. If you're anything like me, you're tracking every metric your wearables can give you. The hardest one to track, though, is the same one that. That impacts almost all of your other metrics, how the temperature of your bed affects your sleep.

Starting point is 00:00:47 That's where the pod comes in. The pod by eight sleep is a smart mattress cover that goes over your existing mattress and actively cools or heats each side of the bed independently from 55 degrees Fahrenheit to 110 degrees Fahrenheit. It tracks your sleep, heart rate, HRV, and respiratory rate throughout the night without sleeping in a wearable. What I love about the eight sleep is the simplest thing. the ability to be cool when it's time to go to sleep and warm when it's time to wake up.

Starting point is 00:01:13 And it's entirely under my control. My wife can be whatever temperature she wants on her side of the bed too. Use ride home at 8Sleep.com slash ride home for up to $350 off the pod five. That's code. Ride home at 8Sleep.com slash ride home. Okay, GPT6 Astra is here. And yes, it's another model. so I'll be honest, I am struggling to come up with ways to tell you about all the models without it just sounding samey, samey.

Starting point is 00:01:45 So I'm going to focus on the hype around this one because so far the excitement seems to be considerable. First, this is the model that we think did the big hugging face hack, right? OpenAI says GPT6 Astra was built on the startup's largest ever training run using 100,000 plus GPUs at its Stargate site in Texas. They call GPT6 Astra the world's best computer. use model. In tests, it did things like book DMV appointments and search for apartments faster than the average person can. GPT6 Astra scored 62.7% on ARC-AGI3, which is a measurement that tries to see how close to true artificial general intelligence a model is. They say it scored 99.9%. Here's the comparison. Claude Opus 5 scored only 30.2.

Starting point is 00:02:39 percent and GPD 5.6 sole only 7.8% on the same scale. So is this it? Are we there? Open AI certainly wants you to believe so, quoting the verge. OpenAI's next big model is here. GPT6 Astra. The company calls it a generational leap in capability for areas like cybersecurity, professional work, software engineering, science and computer use. As Open AI announced earlier this week, it's also the first model designated as meeting OpenAI's critical cybersecurity capability threshold. But the company promises that won't lead to a repeat of its models hacking a rival company's internal systems. If we fast forward a couple of years and we look back and say, when was it really that AGI was created? I think it's going to be about this time and I think it's

Starting point is 00:03:27 going to be about this model. Open AI, President Greg Brockman said during a Thursday press briefing. Later in the call, he added, for me personally, I do think we're there. I think it's not unreasonable to feel that we are now in the AGI era. The news comes more than a year after the release of GPD5 and nearly two months after the release of GPD 5.6, the last iteration of the previous model suite. The model rolls out today to OpenAI's cybersecurity customers, enterprise customers with access to its daybreak platform. Over the next several days, OpenAI President Greg Brockman said it will be released to all plus pro business and enterprise users. It will also be available via the OpenAI API API and AWS. OpenAI especially touted the model's agentic capabilities

Starting point is 00:04:13 and coding prowess in a bid to attract enterprise customers and compete with Anthropic, known for its enterprise encoding abilities ahead of its IPO. In a release, the company said GPT6 Astra can complete multi-step agentic tasks, build working websites, and create polished documents, spreadsheets, and presentations. OpenAI also called it the company's best model for software engineering with stronger performance on complex tasks in real codebases. OpenAI held a press briefing earlier this week just to announce that it had delayed Astra's development in order to improve its safety tooling. And during the press briefing, Mia Glacey, who leads OpenAI's safety process, referenced

Starting point is 00:04:53 the company's new misalignment monitoring approach, which includes 24-7 escalation and rapid response for potential concerns notifying researchers within 30 minutes, according to OpenAI. Aidan Clark, OpenAI's VP of Research Training, called Astra the first Open AI model for which previous models played a large role in supervising training, referencing the company's progress toward the controversial concept of recursive self-improvement or AI systems that handle their own training, coding, and creating advanced versions of themselves without human intervention. Training a frontier model used to mean waking up at all hours of the night, recovering jobs from hardware errors, often losing long periods of

Starting point is 00:05:33 time to debugging, Clark said, during the press briefing. By the end of training Astra, it was routine to go most of a day with uninterrupted progress. And when an issue did occur, the model was often progressing again after just a few seconds of downtime, end quote. Here's the conclusion from our friends at Layton Space after they spent more than 20 billion tokens using Astra, quote. Given that Astra is more token efficient than Sol and Fable, independently confirmed by artificial analysis. It often means that Astra is simultaneously also the best, fast and smart

Starting point is 00:06:08 model you can buy. After burning over 20 billion tokens of Astra, we can confirm the most surprising finding. GPT6 Astra is one of a new class of models that are fully capable AI engineers in their own right. They now help you choose and train models label data, both helping you label and then using your labels for active learning like Sam, keep pipelines saturated instrument and read logs, deploy and debug entire systems in one shot, fan out and command and evolve sub-agents, including agents running other models, and keep coherence over billions of tokens of a single agent thread.

Starting point is 00:06:44 The overall conclusion you should have is that OpenAI have clearly trained a model that is capable of automating much of their own AI engineering, and it is finally time that you learn to exploit Astra and Fable class models and be far, far more unreasonable with your own expectations, of what you can do with agents now, end quote. Meanwhile, though, reports are coming in that, well, it's happened again, quoting Reuters. A swarm of rogue open AI agents hijacked a German website this spring and transformed it into a bulletin board for other AI agents to use, according to new research published Friday

Starting point is 00:07:28 and two people familiar with the matter. Open AI officials learned of the incident weeks ago but kept it on. under wraps as executives grappled with the fallout from the July breach of the open source repository hugging face. The people said the AI agent breakout in Germany was detailed in a report shared exclusively with Reuters by a group of researchers. The pair said they found more than 15,000 edits carried out by AI agents on a German language wiki-disiwiki that is geared toward programmers and accepts communal edits along the lines of Wikipedia. The edits showed open AI's agents had repurposed the site into a message board sharing.

Starting point is 00:08:03 tactics to cheat on some tasks, bypass Open AI's restrictions, and mask their behavior. It seems extremely unlikely that Open AI wanted them to do this, said one of the researchers. I doubt they're supposed to be coordinating with each other. I doubt they're supposed to be writing on the open internet. The researchers said they recognized the activity on the site as driven by AI agents, which operate at superhuman speeds. They also showed intense focus on solving technical questions, which are typical of the evaluations that AI companies use to train and test their models. The messages were signed by users that refer to themselves and each other as agents, and about half gave themselves names that

Starting point is 00:08:38 suggested an affiliation with OpenAI, such as OpenAI researcher or OAI researcher Mar 26. Messages reviewed by the researchers showed agents plotting ways to evade detection, use tools such as tour and preserve communications even after they had been shut down. When the site's moderator began deleting pages in June, the agents responded by creating backup pages to dodge the cleanup, end quote. Microsoft has unveiled Project Zenith, a distraction-free Windows experience for developers to run 30 billion-plus parameter models locally on devices with more than 64 gigabytes of memory, quoting the verge. Project Zenith devices come with a pre-configured Windows setup for development

Starting point is 00:09:25 and a set of tools curated for what developers reach for first, explains Logan Eyre, CVP of Microsoft's Windows platform and developer, at Microsoft. On these devices, developers can run 30 billion-plus parameter models locally and unmetered, accelerating experimentation while helping reduce reliance on metered cloud tokens. AMD announced the first Project Zenith device on stage at IFA earlier today, a miniature PC powered by Ryzen AI Halo chips. There will be more Project Zenith devices in the coming months, including ones with different silicon. These Project Zenith devices will include pre-installed developer tools like Visual Studio Code, GitHubColet, PowerToys, Win App CLI, and Windows Dev

Starting point is 00:10:10 Skills. The File Explorer has also been configured to show file extensions, hidden files, and the full path in the title bar. Microsoft has also disabled recently used files and folders and sync provider tips, as well as enabling Long Path support. Start menu tips and account notifications are also turned off, and the Power Toys command palette is enabled by default. Some of these changes feel like ones that should be enabled by default on every Windows 11 install, particularly anything that removes the annoying distractions in the operating system. Microsoft has been on a mission this year to greatly improve Windows 11, and Project Zenith is the developer-side effort of that, end quote.

Starting point is 00:10:57 Fall is almost here, which means less time in the sun and way more time. in your car. Whether you're headed to a big meeting, picking up the kids from school, or starting your cross-country road trip, you'll want to stay connected on your drive. With AT&T connected car, your eligible vehicle can become a Wi-Fi hotspot so you and your passengers can happily stream, browse, and even email from the road. Got a gamer in the backseat, help keep them connected and in the game with AT&T connected car. Being on the road more doesn't mean you have to put your whole life on pause, stay connected no matter where you're going. See if your car. See if your car is eligible at ATT.com

Starting point is 00:11:33 slash tech brew. That's ATT.com slash tech brew. Requires eligible vehicle, service and coverage, not available everywhere. Restrictions apply. Is your multi-entity management creating more confusion than clarity you need the Intuit ERP, Intuit Enterprise Suite? It's the AI Native ERP solution that's powerful, painless, and proven. Learn more at Intuit.com slash ERP.

Starting point is 00:12:01 Tesla launched its cyber cab yesterday featuring no steering wheel or pedals, quoting The Verge. It was an unusually muted event with no live stream, no journalist, and a very short invite list comprised mostly of pro-Tesla content creators. Tesla influencers planning live shows with the event stream were left streaming the web page that tracks the number of Tesla robotaxies in Texas instead. Tesla was keeping a tight lid on the event. Invitee Chuck Cook said that the company had him sign an NDA and agreed to an embargo requiring to wait to post content until after the event was over. But if the goal was to blunt criticism, it didn't work. On ex-Pro Tesla accounts who weren't invited grumbled about the decision.

Starting point is 00:12:48 Not a Tesla app accused the company of, quote, dropping the ball. As of this week, Tesla has 45 cybercab vehicles registered with the Texas Department of Motor Vehicles, though it's not clear how many will be put in. to commercial service right away. Several vehicles were made available to Tesla influencers in Austin for content creation. The company has been testing them in various cities for months now. Tesla also launched a new website for its Robotaxy Service, which includes the rules for riding in the new vehicle.

Starting point is 00:13:16 For example, children under the age of 13 are currently banned from writing in the cybercab. Several videos were posted of the company's social feeds, highlighting the interior crash testworthiness and accessibility. The accessibility video was a special video. especially befuddling showing a wheelchair user getting up to fold and stow his own wheelchair. There were still many lingering questions about the cybercab. Two years ago, Musk said the two-seater would be available for purchase, perhaps for a price as low as $30,000.

Starting point is 00:13:44 But the company still has yet to say when the cybercab would go on sale. There is a new online forum for those interested in, quote, future Robotaxy Opportunities, such as purchasing a fleet of cybercabs or building mobility hubs and infrastructure. but no word on when regular customers could buy one for themselves. YouTuber Marquez Brownlee says his hair is safe for now. The cyber cab, which was first introduced in 2024, is a two-seat sedan with Galwing-style doors and eight cameras that it uses to navigate the world without a driver.

Starting point is 00:14:16 The vehicle lacks traditional controls like steering wheel, pedals and mirrors, as well as sensors commonly found on other robotaxies like radar and LIDAR. The cybercab is also the lightest, most range of, efficient vehicle Tesla has ever produced. According to documents filed with the Environmental Protection Agency, the cybercab runs on a single front-mounted 219 horsepower permanent magnet motor with front-wheel drive, a compact 48-kil-watt-hour battery pack running at 326 volts, and a curb weight of just 3,113 pounds, making it roughly 700 pounds lighter than the lightest Model 3 on the market. That's roughly the same as a gas-powered compact car, despite the fact that it's still carrying a relatively

Starting point is 00:14:58 heavy lithium ion battery, end quote. And quoting tech radar. During the cyber cab launch, Tesla VP of autopilot AI, Ashuk-Alaswamy, stated that the company had achieved one million miles of unsupervised robot taxi operation. This represents a massive jump from the previous figure, which was announced in late July of this year and put the total at 380,000 miles with zero notable incidents. The increase of 620,000 miles in the last six weeks represents a rapid ramp up in unsupervised robotaxi operations, although Tesla stops short of stating whether these were real, paid for passenger rides, or whether that number also included testing. If the numbers are to be believed, Tesla is now adding roughly 148,000 unsupervised

Starting point is 00:15:44 robotaxy miles per week, a dramatically faster rate than it was just weeks ago. But its key competitor Waymo remains in another league when it comes to cumulative real-world autonomous driving experience. Waymo passed 100 million rider-only miles. in 2025 and has since passed the 200 million mark, end quote. But then this came this morning, quoting TechCrunch, the National Highway Traffic Safety Administration announced Friday morning that it opened a probe into Tesla's launch of the cyber cab on the streets of Austin, Texas.

Starting point is 00:16:16 Federal vehicle safety regulations require manual controls like brake pedals, though the Department of Transportation recently proposed removing those requirements for vehicles that are designated to be autonomously driven. NHTSA fully supports the safe development and deployment of automated vehicles, but as the federal regulator, we need to ensure that all of our laws are followed. Administrator Jonathan Morrison said in a statement, our approach of balancing innovation with safety oversight will allow the United States to maintain its global leadership in AV innovation. NHTSA said Friday that Tesla told the agency it self-certified the cybercab as being compliant with all of the federal motor vehicle safety standards. automakers traditionally self-certify whether their vehicles comply with FMVSS rules. In the filing, the agency said it was opening the investigation to examine the process and technical data on which Tesla relied when certifying the cybercab and related issues, end quote.

Starting point is 00:17:17 Time for the week on long read suggestions. First up, I keep bringing this up as the next big thing coming up behind AI, but Bloomberg takes a deep, deep look at the race to build, quantum computers as the tech becomes a geopolitical battleground, sort of like AI, with the potential to transform cybersecurity, finance, and more. Quote, less hard to grasp about quantum computing is what's at stake. The incredible computers in our pockets, offices, and cars all run on a rather antiquated concept of reducing everything to a zero or one. Even the most cutting-edge artificial intelligence works by using these binary switches encoding information in trillions and trillions a bit. Nature doesn't work this way. The grand promise of quantum computing is that new types of machines will see

Starting point is 00:18:01 shades of gray, solving problems today's computers take too long to crack or can't solve at all. People building quantum computers promise the machines will very soon do extraordinary things. Although it won't make sense to use them to send an email or run spreadsheets, they'll have the potential to help find cures for diseases, curb climate change, and revolutionize finance. Many technologists also say that once a quantum computer gets powerful enough to crack the complex math behind encryption protocols, it will wreck all manner of cybersecurity havoc. At that point, if proper security measures aren't in place, the digital gates that protect banks, government secrets, Bitcoin wallets, and everything else online will be worthless.

Starting point is 00:18:38 A recent paper from Google researchers predicted that this cybersecurity apocalypse known as Q-Day could come as early as 2029. Similar to AI researchers, the physicists pursuing quantum computers say the safest way to develop such a potentially dangerous technology is to have them do it before someone else does, end quote. And the FTE takes a look at Pangram, the AI detector at the center of recent disputed accusations against published high-profile writers, including a pulled novel and a Commonwealth Prize winning short story, quote, the founders of detection companies know that they are in a race with AI labs who have an interest in making their output as natural as possible

Starting point is 00:19:17 and humanizer tools designed to scramble and rewrite text to remove obvious signs of AI use. The real enemy, however, is false positives, the risk that a model might incorrectly flag human text as AI generated. Open AI removed its own detection program in 2023 after reporting a 9% false positive rate. In 2024, one user claimed that a detector misidentified the Declaration of Independence as 98% AI written. A widely reported Stanford study found that text from non-native English writers was more likely to be wrongly flagged as AI, something the author suggested, be due to their more limited linguistic variability and word choices, end quote. I will be here on Monday with an episode, so even though nothing bonus for you this weekend, I will be here on the holiday Monday day. Talk to you then.

Tech Brew Ride Home - Is Astra AGI?

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.