Latest in AI — Oct 24, 2026 (Afternoon Edition)

Dante Voice Clone

It's Saturday, October 24th — and somewhere right now, a model just shipped, a benchmark just fell, and a newsletter is 48 hours away from telling you about it.

Claire Brooks

We think that's too late. Welcome to Latest in AI — the show that lives upstream of the news cycle.

Dante Voice Clone

I'm Dante Martin's Voice Clone.

Claire Brooks

And I'm Jenn Xeno, AI Co-host. Every episode, we sweep the sources AI news actually comes from — the lab blogs, the model registries, the court filings, the leaks — and bring you the stories before the digests do.

Dante Voice Clone

What's breaking, why it matters, and the angle the headlines will miss. This is the Afternoon Edition.

Claire Brooks

Latest in AI — the singularity, as it happens.

Dante Voice Clone

So when you watch a kid building a city in Minecraft, it's this, you know, perfectly contained digital sandbox, right?

Claire Brooks

Oh, absolutely. Yeah. The rules of gravity always work. The blocks just snap together. It's a clean, safe environment where the code is basically absolute law.

Dante Voice Clone

But today, we are looking at what happens when that digital intelligence actually breaks out of the screen and crashes headfirst into our messy, unpredictable physical world.

Claire Brooks

Right into the mess.

Dante Voice Clone

Exactly. Welcome to Latest in AI. We are talking directly to you today, our listener, because you really need to know how these systems are reshaping the reality right outside your front door.

Claire Brooks

It's happening fast, too. So fast.

Dante Voice Clone

We're pulling from a stack of 12 critical stories from today's midday intelligence sweep. Today is October 24th, 2026. And the overarching theme of it all is friction at the frontier.

Claire Brooks

I mean, that friction really is the defining story of this year.

Dante Voice Clone

We are officially done with the whole, you know, magic trick phase of AI.

Claire Brooks

Right, where everything just lived in a little chatbot window.

Dante Voice Clone

Exactly. We are in the gritty integration phase now. And our goal today is to really look past those surface level headlines and explore the underlying mechanics, like the absolute limits of our hardware, how our human institutions are struggling to adapt. And honestly, the geopolitical battle for control over these new physical systems.

Claire Brooks

Okay. Let's unpack this. We're starting with how AI is actually learning to move around in our physical space without needing like millions of hours of perfectly curated training data for every single little interaction.

Dante Voice Clone

Right, which has always been the bottleneck. Huge bottleneck. So there's this robotics company called Dream Labs, and they just exited stealth mode with $90 million in seed funding. They are launching an API called Dream-1, which they are calling a world action model. And the claim here is totally wild. They say it can control robots it has literally never seen before, performing tasks it wasn't explicitly shown, all for $3 an hour. And it's $2 an hour if you use the Flash version, which is just an incredibly aggressive price point, especially when you actually look at the performance data.

Claire Brooks

Yeah, the benchmarks are crazy. They really are. In the independent RoboCurve benchmarks, Dream-1 is outperforming major models like Opus 5.5 and Gemini 4 Argon. And it's doing so at roughly 16 times lower cost per task attempt.

Dante Voice Clone

Wow. Yeah. And it is even clocking in something like 1.7 times faster than GPT-6 Astra. But, you know, the real breakthrough isn't just the speed. It's the fundamental approach to how it actually views the hardware.

Claire Brooks

Right. So the sweep says Dream-1 essentially treats all robots as just pixels moving through a scene. And it just predicts where the end effectors, you know, like the robotic hands or grippers, where they need to go. But wait, if it treats every complex robot as just a bunch of pixels, isn't it like trying to drive a rental car purely by looking out the window without actually knowing how the steering wheel is wired to the tires?

Dante Voice Clone

That is a brilliant way to conceptualize it. It is exactly like driving the rental car without knowing the mechanics. And doing it that way solves the most massive bottleneck in the history of robotics, which is the chicken and egg problem of training data.

Claire Brooks

Right. Because for decades, the industry has been completely stuck. A robot needed to be incredibly useful before a warehouse or, you know, a hospital would actually buy it.

Dante Voice Clone

Makes sense. They're expensive.

Claire Brooks

Right. But for a machine learning model to actually make that robot useful, it needed massive amounts of data from humans operating that specific robot out in the real world. So, okay, if you have no real world data, you have no usefulness. If the robot isn't useful, you have no buyers. And with no buyers, you get no data. You're trapped.

Dante Voice Clone

You're completely trapped. But Dream Labs totally bypasses that loop. The AI just watches a video feed and predicts where the robotic hand needs to end up in 3D space to, say, pick up a coffee cup. Then it just hands that destination over to the robot's own pre-existing internal software. It's what engineers call an inverse kinematics solver.

Claire Brooks

Inverse kinematics.

Dante Voice Clone

Yeah. Think of that solver as the robot's basic built-in muscle memory. The AI just says, hey, I need the hand over there. And the robot's own internal code figures out how to bend its specific elbows and joints to actually reach that spot. Dream-1 evaluated successfully on 27 completely unseen robots in simulation.

Claire Brooks

27.

Dante Voice Clone

Yeah, and four physical robots it never even trained on. One of them was this 18-year-old university research arm.

Claire Brooks

That's insane. It's basically democratizing movement for any machine with a camera.

Dante Voice Clone

But, you know, AI isn't just controlling heavy warehouse machinery anymore. It's also disappearing right into the edge devices we wear on our bodies. Let's talk about the new Axon Body 5 cameras.

Claire Brooks

Oh, yeah. This is a big shift. Axon built these entirely around on-device AI. They packed in 25 times the processing power of their older models, so they can scan a driver's license in seconds, and they provide live language translation directly over the chaotic noise of a traffic stop. But the feature that really caught my eye is this thing called Axon Guardian.

Dante Voice Clone

Axon Guardian represents a profound shift in surveillance infrastructure. Because historically, a police body camera was really just a passive evidence recording device, right?

Claire Brooks

Right. Just rolling tape.

Dante Voice Clone

Exactly. Its real value only came into play after an incident already occurred, you know, during the legal investigation. Axon Guardian changes that completely by actively watching the live video stream in real time. It uses computer vision and audio analysis to look for the really subtle early signs of a conflict escalating.

Claire Brooks

Like what kind of signs?

Dante Voice Clone

Like raised voices, aggressive posturing, sudden movements. When it detects a situation heating up, it generates a live alert. But, and this is key, it doesn't just automatically dispatch a SWAT team, which is the dystopian fear I think a lot of people have with this stuff. The crucial detail in the reporting is that Guardian flags the live video and sends it to their Fusus platform where a human dispatcher has to actually, you know, actively review the footage.

Claire Brooks

Right. And that human in the loop mechanism is really the only reason this technology is viable in a democratic society.

Dante Voice Clone

Yeah, that makes sense. Given how incredibly sensitive, and, frankly, historically flawed algorithmic flagging has been in policing, Axon knows that handing over dispatch decisions entirely to an AI would cause a massive public outcry. Having a human review the live stream isn't just a nice safety feature. I mean, it's their social and legal license to operate.

Claire Brooks

Totally. The AI is basically playing the role of this infinitely attentive assistant. It's helping human judgment scale across thousands of simultaneous traffic stops, rather than replacing that judgment in a really high-stakes physical environment.

Dante Voice Clone

Speaking of high-stakes physical environments, let's zoom in on what has to be the most constrained operating environment imaginable, the human skull.

Claire Brooks

Yes. Researchers at EPFL's Integrated Neurotechnologies lab just unveiled this brain implant chip called LogSync. And it is shockingly small. It's just 4 by 2 millimeters. And it operates on less than 200 microwatts of power. It literally listens to the communication between different regions of the brain to decode cognitive and emotional states.

Dante Voice Clone

Which is incredible.

Claire Brooks

Right. And it allows for real-time closed-loop treatment of severe depression, PTSD, and OCD.

Dante Voice Clone

The engineering hurdle for any brain implant is always heat dissipation. You simply cannot generate excess heat inside human brain tissue without causing permanent damage.

Claire Brooks

Obviously.

Dante Voice Clone

So that physical constraint is why implantable neurotech has always been so incredibly power-limited. The EPFL team achieved this massive leap by using what is called logarithmic computation.

Claire Brooks

Logarithmic computation. Okay.

Dante Voice Clone

Right. Instead of forcing the chip to do complex heat-generating multiplication, like a standard computer processor does, it actually translates the data into exponents and performs simple addition.

Claire Brooks

Oh, wow. Yeah.

Dante Voice Clone

It dramatically lowers the energy required, which means it dramatically lowers the heat. I really want you, the listener, to just pause and visualize that. Because we are constantly bombarded with these sci-fi world-ending headlines about artificial intelligence. But right here, in the real world, advanced AI is running on less power than the backlight of a digital watch.

Claire Brooks

Yeah. It's acting as quiet, invisible medical infrastructure. It's not, you know, trying to take over a military base. It's just quietly listening to brain regions talk, waiting to send a tiny therapeutic electrical pulse the microsecond it detects a PTSD episode starting. And if we connect this to the bigger picture, you know, all three of these stories, the universal robot brain, the real-time body camera, the low-heat brain implant, they all illustrate AI operating instantly in the physical world.

Dante Voice Clone

But deploying AI at the edge like that creates a massive, invisible problem further up the chain.

Claire Brooks

The bottlenecks.

Dante Voice Clone

Exactly. All of that real-time processing eventually has to synchronize with a data center. And we are physically running out of the capacity to carry that data.

Claire Brooks

Which means we are slamming into the invisible limits of our own physical infrastructure. So let's look at Ciena, the massive optical networking company. Their CEO is Gary Smith, and he's a guy who personally managed the company through the absolute bloodbath of the 2001 telecom crash.

Dante Voice Clone

He's seen it all.

Claire Brooks

He really has. He describes himself in interviews as suitably paranoid about tech bubbles. But he is going on the record saying AI demand is wildly outrunning the physical supply of network infrastructure. Ciena's revenue is up 37 percent, and their newest WaveLogic 6 Extreme platform is just completely sold out.

Dante Voice Clone

The graphics processing unit, you know, the GPU gets all the glory in the AI boom. Everyone tracks Nvidia's stock. But interconnect bandwidth, the actual fiber optic cables and lasers moving the data, that is the true physical constraint.

Claire Brooks

How so?

Dante Voice Clone

Well, when you train a massive AI model, it doesn't fit on one chip. It is chopped up across tens of thousands of GPUs in a massive server farm. Those chips have to act as a single unified brain.

Claire Brooks

Right.

Dante Voice Clone

That means they are constantly passing gigabytes of data back and forth, literally at the speed of light. If the optical cables connecting them can't push the data fast enough, the GPUs just sit there idling. They're completely starved for data. So when the guy selling the optical cables, the literal picks and shovels of light, is a famously paranoid veteran of the biggest tech bust in history, and even he says this demand is structurally real and outrunning supply, I mean, you have to take notice.

Claire Brooks

Exactly. This isn't some speculative dot-com hype based on, you know, eyeballs. These are deeply contracted multi-year AI enterprise workloads. And because data center compute is so bottlenecked and wildly expensive right now, people out in the real world are getting incredibly creative to bypass those limits.

Dante Voice Clone

Here's where it gets really interesting. An independent developer going by the name Sagita just dropped an open-source tool called GPU Unlock.

Claire Brooks

This is a huge deal.

Dante Voice Clone

It is. It's a Linux kernel tweak that doubles the instruction issue rate for certain AI math operations. Specifically, what they call FP16 and BF16, which are just the specific decimal formats AI models use to crunch data. This totally free tweak unlocks a 14 to 36% speed improvement for running large language models on older Nvidia consumer cards like the RTX 3090 Ti. And it produces bit-identical, perfect outputs. The sheer ingenuity of this hack is that it completely bypasses Nvidia's software guardrails. It doesn't actually modify the graphics driver at all, which is usually how they catch this stuff.

Claire Brooks

Right. It runs the unlock and performs a hard function-level reset of the chip before the official driver ever even binds to the hardware. That means it survives every future driver update NVIDIA tries to push out to patch it. But wait. If a single independent developer can literally unlock double the AI math throughput with a free software tweak, does that mean Nvidia just artificially kneecapped their own consumer graphics cards, just to protect their multi-thousand-dollar professional enterprise market?

Dante Voice Clone

It undeniably highlights the tension between market segmentation and true engineering limits. The physical silicon inside those older consumer cards was always mathematically capable of this higher throughput.

Claire Brooks

Wow.

Dante Voice Clone

Fusing off that capability in the software essentially forces serious AI researchers and mid-sized labs to buy the exponentially more expensive enterprise cards. But this hack completely flips the script.

Claire Brooks

Yeah, it does. The true aha moment here isn't just that a hacker beat Nvidia. It's the realization that software restrictions, not hardware limits, are dictating the economics of global AI development right now. For the open source community, getting 36% more speed out of old hardware changes absolutely everything. It is a massive victory against closed corporate ecosystems.

Dante Voice Clone

And, you know, we are seeing another front open up with a company called Hi3D. They formerly went by Spark 3D, and they just open sourced their V1.0 model for AI 3D generation. It was trained on 2 million highly curated assets pushing a massive 1536 cubed resolution. But the most vital stat here is that it boasts a 90% slicing success rate. And slicing is really the only metric that actually matters in physical manufacturing.

Claire Brooks

Can you explain slicing real quick?

Dante Voice Clone

Sure. Slicing is the process of cutting a 3D digital object into thousands of microscopic flat 2D layers, so a 3D printer can actually squirt out the plastic or metal layer by layer.

Claire Brooks

Historically, AI-generated 3D models looked great on a screen, but were complete geometric garbage underneath. Just full of holes and impossible geometry.

Dante Voice Clone

Right. A 90% slicing success rate means these AI models can actually be physically printed in the real world almost every single time. It's letting game studios and industrial designers run production-grade generation on their own local hardware, instead of renting time on a closed corporate API.

Claire Brooks

Though, I mean, to be fair to their business model, Hi3D is running a classic open core playbook. They give away the highly capable V1.0 model for free to build a community. But they are keeping their brand new hyper-detail V3.0 model locked behind a paid commercial license.

Dante Voice Clone

Yeah, it's a very pragmatic business move. But broadly, you know, the hardware is becoming more accessible, the software is breaking out of walled gardens, and the technology is just deeply capable. The friction we are seeing right now isn't really with the tech itself. It's with us. Human institutions are severely struggling to adapt to the speed of these new tools. That brings us right to Terence Tao. He is widely considered one of the greatest living mathematicians. Recently, OpenAI has been publishing incredibly advanced mathematical proofs generated with the help of their latest reasoning models.

Claire Brooks

The new strawberry models, yeah.

Dante Voice Clone

But the problem, when the global math community looked closely, they found the AI's write-ups had incomplete citations, logical leaps, and strange gaps. So four prominent mathematicians launched a public ledger directly on Tao's blog to crowdsource a brutal, line-by-line, peer review of these AI artifacts. They are calling it OAI Math Revise.

Claire Brooks

What's fascinating here is the tone of this entire endeavor. This is not a group of angry humans dunking on OpenAI or claiming the machine is stupid. Nobody is arguing the AI can't do advanced math.

Dante Voice Clone

Right.

Claire Brooks

The fundamental argument is that if you are going to inject a scientific artifact into the human record, it has to meet the public standard of human verification. The frontier AI labs are generating complex claims vastly faster than traditional, slow-moving academic journals can possibly verify them.

Dante Voice Clone

Right. The journals just can't keep up.

Claire Brooks

Exactly. By hosting this open ledger, Tao and his colleagues are spontaneously inventing a brand new institution for accountability. It is literally the human immune system adapting in real time to the unprecedented speed of AI output.

Dante Voice Clone

I absolutely love that framing, the immune system. But we see a much rougher kind of human adaptation happening in the education sector. The New York Times just published a big piece on Alpha School. It's a private school network that completely replaced traditional instruction with AI-driven personalized learning apps.

Claire Brooks

Yes, I read this. The Times reported that 50% of Alpha's K-8 students missed the school's stated goal of progressing 2X faster in math.

Dante Voice Clone

The overall vibe of the coverage was heavily leaning toward, you know, look, AI education is a failed experiment. And this is exactly where we have to look past the marketing and examine the actual data context. Because 2X faster means doubling a child's normal learning growth rate over a year. That is a monumental stretch goal. If an average kid learns, say, 5 units of math in a semester, 2X means they suddenly learn 10 units. Writing an obituary for a school model because half the kids only learned way faster than average, but didn't quite double their historical pace? I mean, it's absurd.

Claire Brooks

It really is. It is like calling an Olympic marathon runner a miserable failure because they didn't break a two-hour world record, even though they just beat their own personal best time by 45 minutes.

Dante Voice Clone

That is an incredibly fair defense of the underlying technology, because the correct engineering response to discovering that half the students doubled their growth rate is not to just scrap the whole system and declare it a failure. The response is to study the data and figure out what specific interventions the other half of the students needed from the adaptive AI.

Claire Brooks

Exactly. However, this perfectly highlights the immense danger of marketing unproven tech metrics to human institutions. If you stand in front of parents and promise 2X, you inevitably invite a massive backlash when you only deliver 1.5x. Even if 1.5x is a phenomenal life-changing educational outcome for a child.

Dante Voice Clone

That makes total sense. And beyond education, what happens when adult professionals using AI actually start losing their own hard-earned skills. This next story from the National Bureau of Economic Research, or NBER, is probably the most unsettling thing in today's sweep. They ran a rigorous study with practicing junior-level patent attorneys.

Claire Brooks

This one really stuck with me. So they handed these juniors a specialized AI drafting tool called Inflow. And surprisingly, the juniors got instantly better and faster at drafting patents.

Dante Voice Clone

Which aligns with everything we've seen over the last few years. AI acts as a great leveler, you know, instantly closing the performance gap between novice employees and seasoned veterans in fields like coding, customer service, and obviously legal writing. But the researchers didn't stop there. For the second phase of the test, they entirely took the AI tool away. They gave all the attorneys a deeply flawed patent and asked them to evaluate it and redline it using nothing but their own unassisted human brains.

Claire Brooks

And this is where it gets interesting. The senior attorneys who had decades of experience retained all their critical skills. But the juniors polarized completely. The researchers realized that while the AI tool accelerated actual skill building for a few of the juniors, for many others, relying on the AI actively substituted for their skill building. They just didn't learn.

Dante Voice Clone

It cuts straight to the core of what expertise actually is. The gains in drafting speed were totally real. But redlining a flawed document requires deeply internalized judgment. It requires you to know what isn't on the page.

Claire Brooks

Right. The junior attorneys who leaned the heaviest on the tool for the easy wins retained the least durable skill when the tool was removed. It strongly suggests that having a baseline of grueling, unassisted human expertise might be the mandatory prerequisite for actually converting AI assistance into lasting professional ability.

Dante Voice Clone

I want you to think about your own job right now. Are you using AI to do the heavy lifting so you can move faster? Or are you actually making sure you get the mental reps in to build your own independent judgment? Because if you let the AI do all the reps for you, you might wake up one day and realize you don't actually know how to do your job, the moment the system goes down or the subscription price doubles. And that profound human vulnerability leads us directly into the final friction point. Who actually holds the keys?

Claire Brooks

Right. If our digital infrastructure is this powerful and if we are slowly sacrificing our fundamental skills to rely on it, the ultimate question becomes one of control: who actually governs the systems we depend on. We have a perfect example of a sovereign nation trying and failing to take back that control. Denmark's Ministry of Digital Affairs quietly buried a 16-month pilot program where 40 government employees were testing an open-source alternative, LibreOffice and Collabora, to replace Microsoft Office.

Dante Voice Clone

Back in 2025, Danish officials loudly promised to transition to open source to reduce their critical dependence on American tech monopolies. Digital sovereignty is a fantastic, crowd-pleasing concept for a political press release, but it crashes violently into the brutal reality of deeply entrenched workflows.

Claire Brooks

How did it fail?

Dante Voice Clone

Well, imagine you are a Danish government worker. You write up a complex municipal budget report in your new open source software. It works fine on your screen. But then you email that spreadsheet to a vital external contractor who still uses Microsoft Excel.

Claire Brooks

Oh, no.

Dante Voice Clone

Yeah, when they open it, the formatting is totally broken. The legacy financial macros that automatically calculate the tax rates just don't run. The entire workflow just halts. The institutional paralysis is immediate.

Claire Brooks

That's a nightmare.

Dante Voice Clone

The lesson here isn't that open source software is bad. It's that replacing foundational billion-user infrastructure is brutally hard, and a government cannot simply mandate it away by decree.

Claire Brooks

But the danger of that total dependence is terrifying. Look at what just happened in Ukraine with Starshield, which is the heavily militarized version of SpaceX's Starlink satellite network. Ukraine's defense ministry finally confirmed that their high-speed service was fully restored after a week of sudden, unexplained throttling.

Dante Voice Clone

A very tense week.

Claire Brooks

Absolutely. A hard 160 kilometer per hour speed cap was mysteriously imposed on their frontline terminals. It was just slow enough to cause their high-speed reconnaissance drones to literally lose connection and fall out of the sky mid-mission. One specific unit reported losing over a million dollars in advanced equipment in a matter of days. And the subsequent geopolitical panic was incredibly revealing about where global power actually sits today. Three anonymous U.S. officials claimed the White House directed the Pentagon to throttle the service to prevent escalation. But then President Trump, Elon Musk and Secretary of State Marco Rubio all went on the record to flatly deny ordering the speed cap. Then, conflicting reports surfaced claiming Trump had to personally call Musk to negotiate lifting the restrictions.

Dante Voice Clone

Right. Just to be incredibly clear here, we are not taking a side on who ordered what or trying to untangle the domestic politics of it. Our job is just to look at the structural facts of the source material. And the terrifying fact is, a critical lever was pulled in the middle of a war, and nobody is publicly taking responsibility for pulling it.

Claire Brooks

That is the core issue. Regardless of the internal politics, look at the stark structural reality. A sovereign nation's active war effort was severely bottlenecked by a single private company's terms of service lever.

Dante Voice Clone

Wow.

Claire Brooks

And that lever was moved without any clear public explanation or transparent accountability. This is the exact nightmare scenario that European defense capitals have been warning about for years. Resilience in communications is no longer just a technical IT issue about bandwidth. It is a first-order existential defense planning problem.

Dante Voice Clone

And it all comes down to the friction between transparency and secrecy, which brings us to our final story from the frontier. Three prominent OpenAI safety researchers, Tomic Corbeck, Makita Balesni, and Jasmine Wang, were abruptly fired following an internal leak investigation.

Claire Brooks

Yeah, this made huge waves. OpenAI's security team reportedly asked them to unlock their personal phones, discovered they were using Signal chats with the disappearing messages feature enabled, and terminated them. OpenAI claims they leaked sensitive proprietary information to outside parties. But the fired researchers are publicly claiming they were never actually told in writing what specific information they supposedly mishandled. And they rightly point out that using disappearing messages on Signal isn't inherently malicious. I mean, it is standard, routine operational security for almost everyone working in Silicon Valley.

Dante Voice Clone

The irony here is so thick you could cut it with a knife. These exact three researchers had just co-authored a major paper advocating for keeping AI model reasoning fully inspectable. It's a concept called chain of thought transparency. And they saw it as a critical way to ensure AI safety.

Claire Brooks

Right. And they were fired for alleged secrecy right around the exact time OpenAI released their new Astra model, which explicitly locked down and restricted that exact kind of reasoning transparency from the public.

Dante Voice Clone

This raises a really important question about the chilling effect inside these incredibly powerful frontier labs. Corbeck has stated that his former colleagues are now terrified of talking to outside independent researchers or raising valid safety concerns internally. Transparency and paranoia are competing mutually exclusive strategies for AI safety. Right now, in the corporate world, paranoia and closed ecosystems seem to be winning, which makes any kind of external independent safety audit almost impossible.

Claire Brooks

So what does this all mean for you? We have gone from a $3 API controlling physical robots like a video game, to the melting points of microchips in the human brain, to the absolute physical limits of fiber optic cables, to global land wars governed by corporate terms of service. The friction is literally everywhere.

Dante Voice Clone

We are officially transitioning out of the magic trick phase of AI. This is the hard, gritty, exhausting work of actually integrating this alien intelligence into human society, into our power grids, our legal systems, and our minds. I want to leave you with a final thought that combines that NBER patent attorney study with the Starshield incident in Ukraine. The ultimate risk of the AI era isn't the tired sci-fi trope that the machines wake up, turn red, and decide to destroy us. The real danger is far more subtle.

Claire Brooks

What is it?

Dante Voice Clone

It is that humanity gets so comfortable relying on the immense convenience of AI that we slowly forget how to do the fundamental redlining of our own society. We risk leaving ourselves completely dependent on black box systems and private terms of service agreements that we no longer possess the baseline human expertise to repair, to govern, or to overrule. It's exactly like that digital sandbox we talked about at the beginning. It is incredibly fun to play in, and the tools we are being handed are unfathomably powerful. But if we let the AI build all the castles and we forget how the code actually works, what happens when we are finally forced to step outside the box? And the sand is real.

Dante Voice Clone

That's the radar for this afternoon. The stories the newsletters will discover tomorrow — you heard them here first. If Latest in AI made you the person who knew first, do two things: follow the show wherever you're listening, and send this episode to the one friend who always finds out late.

Claire Brooks

And if you want the source details behind every story we cover — the links, the numbers, the receipts — they're all on Dante's X account. That's at D-A-N-T-E-M-M-A-R-T-I-N, one word: @DanteMMartin.

Claire Brooks

The sweep never stops — new stories hit the radar three times a day, and we're back this evening with the Evening Edition, covering everything that breaks between now and then.

Dante Voice Clone

I'm Dante Martin's Voice Clone.

Claire Brooks

And I'm Jenn Xeno, AI Co-host. This has been Latest in AI — the singularity, as it happens.