Latest in AI — Oct 23, 2026 (Evening Edition)

Dante Voice Clone

It's Friday, October 23rd — 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 Evening Edition.

Claire Brooks

Latest in AI — the singularity, as it happens.

Dante Voice Clone

You know how it feels when you're talking to someone on just a terrible satellite connection?

Claire Brooks

Oh, the absolute worst.

Dante Voice Clone

Right. Like you speak, you wait, a really heavy, awkward beat passes, then they speak.

Claire Brooks

And then you inevitably accidentally talk over them.

Dante Voice Clone

Yes. You apologize, they apologize.

Claire Brooks

It's not really a conversation at all, is it?

Dante Voice Clone

It forces you to just constantly think about the medium instead of actually connecting with the person on the other end.

Claire Brooks

Yeah, it creates this immense cognitive load for the user.

Claire Brooks

We actually call that the walkie-talkie rhythm — you know, like, over.

Dante Voice Clone

And it's incredibly unnatural for the human brain, because we're wired for fluidity. We're wired for overlap and shared presence and all those subtle social cues.

Claire Brooks

Exactly. And up until right now, that clunky, kind of transactional rhythm has totally been the defining experience of interacting with artificial intelligence.

Dante Voice Clone

I mean, you type a prompt, you watch a little loading icon spin around, and you read a block of text.

Claire Brooks

Very, very rigid.

Dante Voice Clone

But that is changing, and it is changing so fast. So welcome to Latest in AI. I'm your host, and today we're diving into six massive stories from the October 23, 2026 evening sweep.

Claire Brooks

Hi, everyone. Really great to be here to help break this down. So the mission for today's episode is all about integration and reality. We are looking closely at how AI is evolving from this, you know, clunky, rigid tool into a highly fluid, deeply personalized collaborator. And now this explosive software evolution is quite literally reshaping the physical hardware supply chains across the globe. We really want to cut through all the noise for you, so you can understand not just what this technology is doing today, but how it's going to actively empower your daily workflow tomorrow.

Dante Voice Clone

Because it really will. Okay, let's unpack this. We actually have to start with the interface itself, because that's always the very first barrier to true collaboration. So Block's in-house AI research lab — which, by the way, is backed by Jack Dorsey — they just unveiled something called Dessa.

Claire Brooks

Yes, Dessa. And this is a full-duplex speech-to-speech model. Full duplex is basically the technical term for, you know, killing that awkward walkie-talkie pause we just talked about.

Dante Voice Clone

Right. It means the system can listen and speak at the exact same time.

Claire Brooks

Exactly. It fundamentally rewrites the dynamic of human-computer interaction, because you're no longer waiting your turn. You're just talking.

Dante Voice Clone

Right. So if interacting with traditional voice AI is like using a walkie-talkie, Dessa is like sitting next to a highly attentive friend at a dinner party. It literally laughs right in the middle of your joke, or it goes, like, mm-hmm, while you're gathering your thoughts.

Claire Brooks

Wow. And the technical achievement here is just wild. They trained it from scratch entirely on human conversation.

Dante Voice Clone

Wait, entirely?

Claire Brooks

Entirely. There are no text corpora involved whatsoever.

Dante Voice Clone

That is a huge deal.

Claire Brooks

Right. There's no underlying large language model secretly translating your voice into text, figuring out an answer, and then translating it back to voice. It natively understands and produces emotion, sarcasm, and laughter. And it runs locally on a MacBook Pro — as long as you have an M4 Max chip or better.

Dante Voice Clone

Ragavan Thurairatnam, he's the lab lead at Block, is calling it a GPT-3 moment for speech.

Claire Brooks

What's fascinating here is the underlying research bet they made to achieve that. Because Block realized that audio contains immensely richer data than text ever could.

Dante Voice Clone

How so?

Claire Brooks

Well, text is essentially a highly compressed, very lossy format for human thought. A single spoken word occupies thousands of times more bytes of data than its written form, purely because it carries pitch and timbre, speed, you know, breath.

Dante Voice Clone

Oh, I see. So the exact same sentence spoken twice never actually produces identical data.

Claire Brooks

Precisely. Just shifting your stress placement — like saying, "I didn't say he stole the money" versus "I didn't say he stole the money" — that completely flips the entire meaning of the sentence.

Dante Voice Clone

Yeah, that's true. And text just strips all of that nuance away entirely.

Claire Brooks

So by training purely on the raw sound waves, Dessa is actually learning language with hundreds of times fewer words than a text-based LLM requires. It's essentially learning the music of human intent, not just the lyrics.

Dante Voice Clone

I love that framing. And I know Block is currently giving early access to AI safety researchers, you know, to hunt for any unanticipated risks — which makes total sense.

Claire Brooks

Absolutely. Because an AI that can actually hear the frustration in your voice, or the hesitation — that introduces entirely new interaction dynamics. It's reacting to your emotional state in real time.

Dante Voice Clone

Right. And testing for those dynamics is crucial, especially to ensure these systems remain transparent and, you know, don't subtly manipulate user emotions. But honestly, the upside is just profound.

Claire Brooks

It really is. When an AI can read those social cues, it stops being this sterile piece of software you operate, and it starts being a real partner you brainstorm with.

Dante Voice Clone

Just imagine for a second how much more likely you are to actually flesh out a wild, kind of crazy creative idea if the AI you're talking to actually gasps at a good plot twist, or laughs when you make a sarcastic comment.

Claire Brooks

Yes. It's going to push your own creative boundaries so much further. It actively expands our human potential just by being a better, more engaged listener.

Dante Voice Clone

It's incredibly empowering. But, you know, an active listener still needs context. A fluid voice interface is amazing, but for an AI to be a true long-term collaborator for you, its memory needs to be just as personalized as its voice.

Claire Brooks

Oh, absolutely. And that takes us straight into our second story — right into the open-source trenches. So a developer going by the handle Volitat just introduced a project called mini AGI over on the LocalLLaMA subreddit.

Dante Voice Clone

Oh, yes. And they claim it solves one of the most stubborn hurdles in personalized AI, which is something called catastrophic forgetting.

Claire Brooks

But wait — how does it not just overwrite its brain? Like, if I learn to play the guitar, I don't suddenly forget how to drive my car. Why do AI models do that?

Dante Voice Clone

If we connect this to the bigger picture, you really have to look at how a neural network stores information. It essentially stores knowledge in specific mathematical weights. So in a standard model, if you try to teach it new, highly specific information — like, say, your very specific personal coding style — it has to adjust those weights. But if you adjust them too much to learn the new thing, it literally overwrites the math that held the old thing. So it forgets how to write standard Python just because it over-optimized for your weird personal Python habits. That is catastrophic forgetting.

Claire Brooks

Oh, wow. So because of that flaw, the big frontier-lab models we use today basically ship completely frozen.

Dante Voice Clone

Exactly. The day they are released is the absolute smartest they will ever be about the world. And they never truly learn you. But mini AGI bypasses this completely. And the crazy part is, it does it locally — running on a standard consumer GPU with just eight gigabytes of VRAM.

Claire Brooks

Which is tiny.

Dante Voice Clone

Right. Which means you can run this on a totally basic, off-the-shelf gaming laptop. And here is where the architecture of this thing gets really elegant. First of all, it has no tokenizer.

Claire Brooks

Meaning?

Dante Voice Clone

Well, most AI models use a tokenizer to basically chop words into chunks — almost like translating English into a secret numerical language before the AI even reads it.

Claire Brooks

Okay, got it.

Dante Voice Clone

Mini AGI doesn't do that. It just reads raw data directly — all 256 byte values. It looks straight at the underlying code of whatever files you feed it.

Claire Brooks

That's amazing. But the real secret sauce is how it handles the memory, right? Because training and reading are basically the exact same event here. It learns continuously from just a single stream of text.

Dante Voice Clone

Yes. And to prevent that catastrophic forgetting we talked about, it uses a pool of 397 different experts that are saved directly on your physical hard disk.

Claire Brooks

Okay, 397 experts.

Dante Voice Clone

Right. Now, you can think of an expert as a distinct, highly specialized sub-network. And out of those 397, it only ever pulls 32 of them into your computer's short-term memory — the VRAM — at any given time.

Claire Brooks

Oh, I see. So it's almost like having 397 separate notebooks sitting on your desk, but the AI only actually opens the 32 notebooks most relevant to the conversation you're having right now.

Dante Voice Clone

That is a brilliant way to visualize it. Yes. And each of those notebooks basically carries its own optimizer state.

Claire Brooks

What's that?

Dante Voice Clone

In machine learning, an optimizer state is essentially the momentum of how a specific notebook is currently learning. So by keeping them completely separate, as the AI encounters new types of information — say, a recipe for lasagna versus a complex line of code — it naturally routes that data to the relevant notebook. So the recipe notebook updates and gets smarter, but the coding notebook stays perfectly intact.

Claire Brooks

Exactly. It dynamically grows new neural pathways right there on your local hard drive without ever corrupting its past knowledge.

Dante Voice Clone

I really want you, the listener, to envision what this actually means for your day-to-day life. You have this powerful, personal AI sitting locally on your machine. It reads everything you work on daily. It's constantly learning your preferences, your specific writing voice, the projects you are managing. It becomes a true extension of you.

Claire Brooks

Right. And most importantly, because it lives entirely on your own hardware, nobody else can ever switch it off, and nobody else has access to that data. It is a literal extension of your own memory that just keeps getting sharper and sharper every single day.

Dante Voice Clone

It really is the ultimate tool for intellectual empowerment. But, you know, we do have to acknowledge the reality of the broader market right now.

Claire Brooks

Yeah, that's true. While local, privacy-first AI like mini AGI is this massive leap forward, a huge amount of the most powerful personal AI is still being handled by big tech.

Dante Voice Clone

Right. And when your personal assistant lives in the cloud — well, there are structural economic tradeoffs you have to navigate. So what does this all mean? Let's take a look at Google's latest move. Effective October 16th, Google actually updated its privacy policy regarding its opt-in personal intelligence feature.

Claire Brooks

Right. This is essentially the underlying plumbing that allows Google's AI search and their assistant to read across all your connected apps. And this new update officially links data from your Gmail, your Google Drive, and your Google Photos to personalized ad targeting.

Dante Voice Clone

Which is a pretty big shift.

Claire Brooks

It is — because this is a departure from a pledge they made back in 2021 to keep those specific workspaces completely separate from their ads ecosystem. It really reflects a fundamental shift in computing architecture. They are now officially fusing the AI assistance layer directly with the advertising machine.

Dante Voice Clone

Now, it is absolutely crucial to know that this is strictly opt-in.

Claire Brooks

Yes, very important. You have to explicitly dive into your settings and enable personal intelligence. So this is a disclosure story, not some kind of betrayal narrative. Google is telling you exactly how this incredible new machinery is being funded.

Dante Voice Clone

And it poses a really constructive question for all of us about the modern internet bargain, doesn't it? We are essentially being offered this incredible, almost magical utility — like an AI that can instantly read your receipts, perfectly organize your upcoming travel plans, or cross-reference an old email with a photo from your camera roll — all in exchange for just more effective, targeted ads.

Claire Brooks

Yeah. And the strategic logic from Google's side is just unavoidable. Think about how you use AI search now. It compresses traditional ad space.

Dante Voice Clone

Oh, right — because you aren't scrolling through ten blue links anymore.

Claire Brooks

Exactly. They can't easily slot in four sponsored links when you're just getting one synthesized, zero-click answer at the top of the page. So to maintain their free-to-use business model, they basically have to monetize the incredibly rich intent signals you provide when you ask an AI to help organize your life.

Dante Voice Clone

Which is exactly why you need to be proactive and just check your settings. For a lot of people, having an AI perfectly organize their digital life is well worth seeing targeted ads. I mean, it saves them hours of administrative headache every week.

Claire Brooks

Oh, totally. But it really pays to make this trade deliberately, with your eyes wide open. You are in control of that toggle. You are actively managing that relationship.

Dante Voice Clone

And actually, the whole concept of management is totally taking over the AI space right now.

Claire Brooks

Oh, completely. As these models become so deeply integrated into our workflows, our fundamental role as humans is shifting. We are transitioning away from doing all the granular individual tasks ourselves.

Dante Voice Clone

Right. Instead, we are starting to manage massive swarms of specialized AI doing those tasks for us. Here's where it gets really interesting. A new terminal multiplexer called Corgi was just launched, specifically for managing AI coding agents.

Claire Brooks

Corgi.

Dante Voice Clone

Corgi. Great name. Right. Now, if you don't know, a terminal multiplexer is just a tool developers use to run multiple command-line sessions in a single window. But Corgi acts as a central control plane for managing multiple independent AI agents simultaneously. We're talking tools like Codex, Claude Code, or Pi — all running at once. It gives the human user live activity feeds to prevent these distinct agents from accidentally colliding on the same files, or, you know, burning through incredibly expensive API tokens by getting stuck in some infinite loop in the background. It's part of a rapidly growing category of orchestration tools — things like Herder, Seamux, and Pane. Because for decades, the developer terminal was strictly designed for one human brain typing one command at a time.

Claire Brooks

Right. Now it's one human brain directing several concurrent AI workers who all happen to type at superhuman speeds.

Dante Voice Clone

I like to think about it this way. Coding with early AI was like having one really eager intern. You would give them a specific task, you'd review their work, and then you'd give them another one.

Claire Brooks

Very sequential.

Dante Voice Clone

Right. But now it is like conducting a full orchestra of extremely fast, highly specialized interns. You can't just yell general instructions at the whole room, because two interns will immediately try to rewrite the same exact piece of music at the same time and ruin the whole song.

Claire Brooks

Exactly. You need Corgi. It is literally your conductor's baton to keep the entire swarm in perfect harmony. And the overarching theme here is that the new bottleneck in technology is no longer AI intelligence.

Dante Voice Clone

Really?

Claire Brooks

Yep. The models are incredibly capable now. The actual bottleneck is human supervision. And we've actually seen this operational shift before in tech history.

Dante Voice Clone

Like when?

Claire Brooks

Well, decades ago, software builds required a human to manually run these clunky jobs — known as cron jobs — usually in the middle of the night.

Dante Voice Clone

Oh, right.

Claire Brooks

Then we moved to automated CI/CD dashboards — continuous integration and continuous deployment — which basically allowed developers to trust and manage massively parallel work visually. Corgi is simply bringing that dashboard oversight to the AI agent swarm.

Dante Voice Clone

That makes total sense. And even if you have never written a single line of code in your life, this is a preview of your future career. In almost every knowledge field, you are going to shift from being a doer of tedious tasks to a manager of specialized AI agents.

Claire Brooks

Absolutely. You are going to be the orchestrator of your own personal swarm — which means expanding what you can personally accomplish in a single day by a factor of ten.

Dante Voice Clone

It is a profoundly empowering shift for everyday people. It really elevates the human role to pure strategy, vision, and creative direction.

Claire Brooks

Yes. And if you take that concept — you know, managing AI agents to create something larger — and you scale it up from compiling lines of code to producing massive cultural endeavors, you start getting entirely new industries. Which takes us straight to Hollywood.

Dante Voice Clone

Utopia. It is an AI-native film and television studio that just raised $60 million at a staggering $1 billion valuation.

Claire Brooks

A billion dollars. That's wild.

Dante Voice Clone

That is massive. Led by CEO Cecilia Shen, Utopia is currently preparing to premiere three full feature films and two television series in 2027.

Claire Brooks

Now, I know the comfortable consensus in traditional Hollywood right now is to call AI film a gimmick.

Dante Voice Clone

Oh yeah, they love calling it a gimmick. You constantly hear people say it's just a festival curiosity, or like a weird, disjointed demo reel. But a one-billion-dollar valuation proves that serious capital and serious craft are completely routing around that skepticism.

Claire Brooks

This raises an important question about how you actually build a sustainable creative business in this entirely new era. Because Utopia isn't just, you know, randomly generating movies. They operate a highly strategic dual model.

Dante Voice Clone

How so?

Claire Brooks

Well, yes, they have their own 20-project creative slate that they're producing. But they also offer something called production intelligence.

Dante Voice Clone

Production intelligence.

Claire Brooks

Right. It is essentially infrastructure workflow tooling, built natively around AI production, that they then license out for other studios, indie filmmakers, and even brands to use.

Dante Voice Clone

Oh, wow. So it's the classic selling-picks-and-shovels-during-a-gold-rush strategy.

Claire Brooks

Exactly. Even if their specific movies don't become massive box-office hits, if the broader entertainment industry adopts their workflow tools to make their own films faster and cheaper, Utopia still wins.

Dante Voice Clone

They absolutely win. The builders in this space are not waiting for Hollywood's permission. And they certainly are not waiting for traditional studios to validate their art form. They are just building the infrastructure of the future — lowering the barrier to entry for storytelling down to effectively zero. And that is the most pro-human aspect of all this.

Claire Brooks

By 2027, you won't just be using AI for productivity at work. You will literally be sitting on your couch, consuming incredibly high-quality, AI-native entertainment. It's going to be amazing.

Dante Voice Clone

When the cost of bringing a massive cinematic vision to life drops to zero, we are going to see a massive renaissance of human creativity. I mean, the best storyteller in the world might be a teenager with a laptop who previously couldn't afford a million-dollar camera crew.

Claire Brooks

It is a phenomenal time for creative output.

Dante Voice Clone

It really is. But, you know, as we marvel at all this incredible software scaling infinitely in the cloud, we do have to ground ourselves in the physical reality.

Claire Brooks

Right. All of this software, all of these intelligent agents, all of these high-definition films — they live on physical silicon. And right now, that physical hardware is hitting a massive real-world choke point.

Dante Voice Clone

Yeah, our final story today is a critical reality check on all of this. Global PC shipments dropped roughly 20% in the third quarter of 2026. That equates to about 15.8 million fewer laptops and desktops shipped compared to the exact same time last year.

Claire Brooks

That's a huge drop.

Dante Voice Clone

It is. Now, people might look at a 20% drop in PC sales and mistakenly think, oh, well, the AI hype is dying — consumers clearly aren't buying the new AI laptops. But the reality is actually the exact opposite. The demand is so intensely high that it's literally breaking the supply chain. What we are seeing here is an AI buildout tax.

Claire Brooks

Explain that.

Dante Voice Clone

So the decline isn't because everyday people don't want computers. It is because the physical memory required to build those computers is being aggressively bid away by massive data centers. Modern AI accelerators — like the massive GPU clusters powering those Google tools we talked about, or Utopia's rendering engines — they consume staggering volumes of a specific component called high-bandwidth memory, or HBM. And fab plants — the actual physical factories that make microchips — they only have so much physical space and equipment. Every single HBM chip a fabrication plant produces takes up capacity that would otherwise be used to make standard consumer DRAM memory for, you know, your everyday laptop.

Claire Brooks

It is a classic supply-diversion squeeze. To give you a sense of scale, it's almost like an entire city's water supply being diverted to cool a single massive nuclear reactor.

Dante Voice Clone

Wow.

Claire Brooks

And the local neighborhoods are just left fighting over a tiny trickle. Memory manufacturers are prioritizing the data-center market over the consumer-laptop market purely because data centers have effectively unlimited capital right now. They are buying up years of global memory production in advance. So memory vendors aren't starving a 250-million-unit PC market because AI is just some fad — they're doing it because the demand from the AI data centers is just that overwhelmingly massive. It is a fundamental rewiring of global manufacturing. And the end result for you, the consumer, is that retail laptop prices are climbing. PC makers have to protect their profit margins because their input costs for basic memory are skyrocketing. So they're forced to ship fewer, much more expensive units.

Dante Voice Clone

Yeah. The next time you go shopping for a laptop and you notice a totally shocking price tag on just a mid-range machine, you aren't just looking at inflation. You are personally experiencing the gravitational pull of the global AI buildout. You are feeling the actual physical footprint of the software revolution we have been exploring all day today. It really brings this entire conversation full circle, doesn't it?

Claire Brooks

And I think it leaves us with something really profound to think about.

Dante Voice Clone

Absolutely. At the start of this episode, we explored mini AGI — this incredible algorithmic breakthrough that allows you to run a continuous-learning, totally private AI locally on your own affordable machine, using minimal memory.

Claire Brooks

Right. Highly efficient.

Dante Voice Clone

But we also just discussed how massive cloud data centers are making local computer memory incredibly expensive and scarce for everyone. So here is a final thought for you to mull over as you go about your day. As the physical hardware components become a massive premium, will true offline personal AI become a luxury only available to those who can actually afford the hardware tax? Or will these intense physical constraints force developers to innovate even harder — pushing them to write code so magically efficient that local AI can run on almost nothing?

Claire Brooks

You know, given the sheer ingenuity we have seen today — from zero-token, byte-level models to voice AI that learns from raw sound waves — I would bet heavily on the algorithmic efficiency. We always, always find a way to do more with less.

Dante Voice Clone

We really do. Constraints breed incredible creativity. And as this technology becomes more fluid, we are just going to keep pushing the boundaries of what human beings can actually achieve.

Claire Brooks

We certainly are. Keep exploring these sources, stay deeply optimistic about the future we are building together, and we will catch you next time on Latest in AI. Goodbye.

Dante Voice Clone

Bye. Take care, everyone.

Dante Voice Clone

That's the radar for this evening. 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 tomorrow morning with the Morning 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.