Sabrina Ortiz

For years, voice dictation's fatal flaw was that it struggled to accurately understand what you were saying. While AI has nearly eradicated that issue with products such as Wispr Flow and Google Rambler, a new problem has taken its place.

Just a couple of weeks ago, Google debuted its Pixel 11 lineup, and one of the standout AI features was the new Rambler voice dictation, which integrated into Google Keyboard (Gboard) and offered the same capabilities that Wispr Flow has been able to do for years: precisely understanding users' voice dictation, eliminating filler words, and following a user's train of thought.

As someone who cannot work without Wispr Flow, I tested out Rambler for all my communication, both personal and work, as soon as I got my review units. I soon ran into a hang-up I had never noticed before: tone.

To my boyfriend's credit, he was the first to bring it up, saying he preferred Wispr Flow because it adapted better to his expressions, such as automatically including exclamation points. In what would become foreshadowing, I told him to give Rambler some grace, since it, like Wispr Flow, claims to get to know you better over time and eventually implement that in its dictation.

However, despite my consistent use, Rambler has yet to adapt to my excited and bubbly tone, opting instead to send the driest texts. The catalyst for this review was when I was Slack messaging TDV editor Jason Hiner and had to clarify that I had used Rambler because the message I sent sounded absolutely nothing like me. It was much less friendly, more serious, and more to-the-point than I naturally speak or write in Slack messages.

While it's still polite, the absence of tone indicators, such as punctuation, that you typically use can make people who know you think something is wrong.

For instance, my boyfriend sent a text via voice dictation using Rambler that was fine on its face, but the punctuation made it read as dry and dismissive. After a long day, I misread it that way and brought it up, sparking an issue that never would've happened without the dictation software's punctuation choices. There's actually research to back this reaction:

• A 2016 Binghamton University study found that texts ending with a period were rated less sincere than those that did not.
• A 2025 follow-up study found that when periods were placed after every word, for instance, typing "What. Do. You. Need." instead of "what do you need," the perceived "mean" rating, or how mean of a tone the reader perceived in the texts, was higher, highlighting how readers interpret punctuation in a sentence as an intentional, meaningful act.
• A 2018 The Atlantic article explored the phenomenon of people using so many exclamation points while texting and featured Gretchen McCulloch, a linguist who studies online communication, who said: The single exclamation mark is being used not as an intensity marker, but as a sincerity marker. If I end an email with ‘Thanks!,’ I’m not shouting or being particularly enthusiastic; I’m just trying to convey that I’m sincerely thankful, and I’m saying it with a bit of a social smile.

Our Deeper View

In theory, AI voice dictation features are huge game changers for productivity. If you say it out loud, it will transcribe it for you, saving you the time it takes to type, since most people can speak at least 2-3x faster than they can type. But transcribing what you say is only half the battle, as the consequences of not having that so-called social smile can be serious. For that reason, Wispr Flow remains the undisputed leader until Google fixes this in Rambler. However, this discussion does raise the question of how AI will transform text communication in the future, and how our perceptions of tone in text will evolve as a result. Oh, and to prove my point: I used Wispr Flow to write this entire article, and it understood my intent and intonation perfectly.

Mac has been the undisputed home of local AI, but Nvidia is giving Microsoft an assist to get Windows in the game.

On Thursday at IFA 2026, Nvidia unveiled new tools and hardware aimed at making it easier for more PC users to run AI locally on Nvidia GPUs for faster local inference. For starters, Nvidia RTX Spark arrives in October in new Windows PCs from Lenovo and Acer, on display this week at IFA. RTX Spark has a 1 Petaflop RTX Blackwell GPU, up to 128GB of unified memory and a highly efficient 20-core Grace CPU, which, when combined with new agent frameworks, gives Windows PCs the power to run always-on AI agents locally.

Nvidia is also offering a simpler model setup on Windows for three of the most popular agent apps: Perplexity Portable Computer, Hermes Agent, and OpenClaw. Here is a quick rundown of the features, according to the release:

• Perplexity Portable Computer: Will be available on Nvidia RTX GPUs with at least 24GB VRAM running Linux or Windows
• Hermes Agent: Coming soon, configuring a local model in Hermes will be streamlined with one-click setup across RTX and DGX systems on both Windows and Linux
• OpenClaw: Nvidia worked with Microsoft to reduce set-up friction and the result is that the OpenClaw Windows App simplifies setting up an optimized local model on any RTX GPU with at least 24GB of VRAM

Building on its efforts to improve AI use efficiency, the tech giant also unveiled NVIDIA Personal AI Router (PAIR), a free, open-source software tool that can coordinate a household's PCs to run local AI together. NVIDIA says that PAIR can automatically discover compatible PCs on a local network and route independent inference requests to the system with available capacity. Ultimately, this is meant to bypass the bottleneck that is caused when multiple agents or tasks are waiting on a single GPU.

Our Deeper View

The rise of AI agents has led working professionals to discover brand-new ways AI can assist with their work. However, the caveat is that agents can quickly rack up token costs and so power users typically want to move to running AI locally. Until now, it has largely meant a reliance on Mac products, as seen by the shortage of the Mac mini and Mac Studio, and the rollout of cutting-edge AI features arriving on Mac first. The release of Nvidia RTX Spark on Windows by Nvidia is significant, as it gives users who don't want to be locked into Apple's walled garden more choice. It may also sway developers from overlooking Windows when unveiling the latest features on desktop apps, which would be a win for Microsoft and for Windows users.

Disclaimer: Sabrina Ortiz's travel to IFA 2026 was paid for by IFA. The Deep View's coverage is editorially independent from the companies we cover.

HTC, once a dominant force in smartphones, is throwing its weight behind smart glasses.

On Tuesday, HTC made its Vive Eagle AI glasses available in the US, Europe, and Australia, expanding from their original spring launch, when they were available only in Taiwan. The glasses offer exactly what you'd expect: AI assistance, speakers, microphones, and a camera. However, there are two key distinguishing features that users, and other manufacturers, should note.

Before jumping into that, however, here's a rapid-fire list of specs and my thoughts on them:

• Camera: 3K video quality and 12 MP camera; good quality captures most of the time, though sometimes overexposed or blurry in places
• Weight: 49 grams, lightweight, comfortable and on par with Meta Ray-Bans
• Lens: Zeiss lens in three options; also prescription compatible, -8D to +4D
• Speakers: Two stereo, bass-enhanced, open-ear speakers; sound is great, though there is the expected sound leakage
• Mics: 4-mic array, 1 directional (in the nose bridge) and 3 omnidirectional, noise cancellation software; in a loud cafe, my mom could still hear me well on a phone call
• Processor: Snapdragon AR1 Gen 1; same chipset as Meta Ray-Bans, performs on par
• Battery: HTC claims 4.5 hours of music, 3 or more hours of calls, and 36 or more hours of standby; I have only used it session-based, so I can't verify what all-day use is like

I intentionally left the privacy features outside the quick rundown, as they deserve their own section. The company is positioning the product as "HTC's privacy-first AI smart glasses," and in my testing, it more than delivered.

As soon as you cover the LED light while recording, it stops capturing media and verbally says, "LED covered, recording has stopped," to prevent bad actors from recording non-consenting individuals. While the feature is offered on other models, including Meta Ray-Bans, I found this to be the most responsive so far.

The camera is also disabled when the glasses are not worn, a feature I only noticed after trying to take a selfie with them. This is meant to prevent individuals from being recorded by the glasses lying inconspicuously on a table. Lastly, they are ISO 27001-certified for information security management and ISO 27701-certified for privacy information management, certifications not held by other AI glasses.

The other standout is the magnetic Powerboost accessory, roughly the size of a flash drive, which lets you charge your smart glasses while still wearing them. This solves the problem of dead batteries that require you to remove the glasses to use a case or cable. Downside: the PowerBoost accessory can't fully charge the glasses, even when fully charged itself, likely because it's a 120 mAh battery. It boosted the battery from 39% to 84% in about an hour before dying in my tests. It's still useful for quick top-offs to get you through a full day, which is especially useful if you have prescription lenses and want the glasses to be your daily drivers. I'd love to see more companies adopt this approach.

I saved the AI features for last because I found them about as useful as competitors like Meta: cool in concept, but not entirely life-changing yet. While HTC does give users the option to choose between Gemini and ChatGPT, which is notable, the company notes that the experience may differ from using the models on your phone or other devices. I found that to be entirely true, with the answers I got using the Gemini and ChatGPT apps being much more robust and preferable.

Our Deeper View

The VIVE Eagle will launch at $499, and the Power Boost accessory is an extra $49, placing them at the high end of the spectrum, nearly double the price of the standard Meta Ray-Bans. However, if privacy is a priority, it may be worth the premium. Unlike Meta and Google, HTC is not a company that makes money off your data, so that also makes it a safer long-term bet if you prioritize privacy. The other big reason to buy is if you want smart glasses to be your full-time glasses and don't want to take them off to charge them. In addition to the PowerBoost accessory, you can also simply charge the glasses while wearing them because of the magnetic charger on the arm. While Gemini and ChatGPT aren't a game-changer on the glasses yet, they're likely to improve and could be more appealing as a long-term bet on the glasses.

Sonos built one of the most coveted home sound systems. Now it's opening that platform up to AI in a unique way.

On Tuesday, Sonos introduced two new flagship products, the Sonos Beam Ultra and Sonos Ace Ultra, as well as the latest version of its audio operating system, Sonos 27, which is infused with AI experiences. However, Sonos is differentiating itself from most companies by taking an open approach that makes using AI to manage its device ecosystem more useful and compelling.

With the launch of Sonos 27mcp, users can connect any external AI assistant or agent, including ChatGPT or Claude, to their system to perform tasks such as playing music or controlling what they are listening to. The biggest benefit here is that users can control their smart home audio right from the conversations they are already having with their favorite assistant, on their phone or computer.

This is a unique approach compared to competitors who often lock users into their AI systems and spend significant resources trying to build the best AI. However, the AI labs have already built highly capable models, and users have often already picked their favorite. While it may keep users from relying on the Sonos 27voice, the new assistant built for music and mood curation creates a more seamless experience with the products, and that's Sonos's ultimate goal.

"We actually want customers to control Sonos; we want customers to have great experiences in their home, and we see the future of control being really much more open in access," Andrew Sutherland, senior director of software product management at Sonos, told The Deep View.

Sonos also previewed where it plans to go next with AI: Sonos Custom Agents. This experience will allow users to create their own agents, each with its own model, including third-party models if the user chooses, and to name them individually.

"Think about the power of having a single speaker, where you can have access up to like 10 different agents if they have different tasks across your home or they have different purposes, and through Sonos, you can set up bringing your LLM of choice or your model of choice, and then you can access each one of those with different types of personas and voice," added Sutherland.

Sonos 27 will become available through a software update to all Sonos S2 products, with some features only available on certain products. The next-generation flagship products, the Beam Ultra and Sonos Ace Ultra, are available for pre-order starting today.

Our Deeper View

One of the original and most popular uses of voice assistants over the past decade was for controlling your smart home, with Amazon's Alexa being the leading example. However, those assistants have become antiquated compared to today's LLMs, despite efforts to catch up. As a result, Sono's approach feels like a winning play. It's finding ways to incorporate leading AI models into its products so that users can manage their devices with one of the latest voice assistants without having to context-switch to an inferior or outdated assistant.

AI is best known for transforming coding. Now, one startup is making the case for physics.

On Tuesday, the AI-native physics research lab, Physical Superintelligence (PSI), emerged from stealth with $58 million in seed funding, led by Breakthrough Energy. It is launching with two proofs of concept:

1. A productized piece of its core platform, Emmy
2. Joining as a founding technical partner for the Fermi Explorer Mission, a nonprofit organizing the first privately funded interstellar space mission and the first AI-planned probe to Alpha Centauri

Emmy, named for renowned physicist Amalie Emmy Noether, combines PSI's reasoning engine, consisting of sovereign pre-trained and post-trained models, with a large curated inventory of simulations to tackle research problems at a pace much quicker than humans could, according to the company. Moreover, Emmy can reason through a problem, then test its conclusions until its findings are verifiable, as Matt Pines, co-founder and CEO, told The Deep View.

"Our systems run research campaigns: they decompose a problem, generate candidate approaches, and test them against simulation, live measurement, or machine-checked proof," said Pines. "Nothing counts as a result until it survives a check that sits outside the model."

Initially, a subset of Emmy's capabilities will be used for optimizing terrestrial and orbital AI data centers and factories. PSI has already signed commercial agreements and live pilot deployments on operating data center infrastructure today, according to Pines.

The second prong of the launch is PSI's involvement in the Fermi Explorer mission, whose ultimate goal is to launch the first spacecraft to another star system, targeting Alpha Centauri, the closest star system to Earth. This initiative is a major undertaking because Alpha Centauri is roughly 4.37 light-years away, which would take about 80,000 years to reach from Earth at the speeds of current spacecraft. That makes it quite a feat of engineering to build a vessel capable of the journey.

PSI has already claimed to have contributed to the mission by validating its physics and identifying a substantially more efficient trajectory within the mission’s mass and budget constraints. The company is using this finding to demonstrate that a small team using AI-native physics could do the work typically required of a national laboratory. This reflects the company's broader mission to contribute to discoveries that are both commercially and scientifically valuable.

"Fermi asked us to assess mission feasibility: the propulsion, trajectory, and power questions that determine whether the mission closes," said Pines. "Our technology ran the analysis, with our physicists directing the work. Fermi's technical team, which comes out of Starcloud, verified the analysis. The report was also written so the analysis can be rerun, and reproduction is the standard we want to be held to."

PSI was founded by Pines, Alex Klokus, and Dr. Alexander D. Wissner-Gross, Ph.D, who combined to bring expertise across physics, economics, government, and tech. The broader team comprises physicists, AI researchers, experimenters, and builders, and PSI is actively hiring more talent. Interested applicants can apply online.

Our Deeper View

ChatGPT became the catalyst for the current AI boom, and since then, we have seen many companies try to compete by creating AI products. The result is that many of these products end up being repetitive or AI-washed offerings that have largely caused mainstream AI fatigue. However, some labs are developing focused, task-based AI solutions to solve big problems. Physical Superintelligence is a prime example, as it showcases just how instrumental AI can be as a catalyst to spur further innovation and development, even unlocking discoveries that have been very difficult to solve, with this extreme example of building a vessel capable of reaching Alpha Centauri. It's refreshing to see teams with ambitions this big.

Manufacturers have unlocked a simple formula to capitalize on the AI hardware craze: take ordinary items, incorporate microphones, and layer in an AI assistant.

The promise is an AI tool that escapes the traditional screen and accompanies you everywhere. While these products often do that, the issue is that their main features are often duplicative, overlooking the fact that AI assistants are in a device that's always with you. On your phone, you already have a voice recorder that, when combined with an arsenal of apps that can transcribe, analyze, and answer questions, can already deliver what most of these AI devices do.

A perfect example is a product I recently tried called the Flowtica Scribe, which calls itself "the world's first AI pen." If you are like my roommate and excitedly assumed an AI pen would do something groundbreaking, like digitally transcribe the words you write with it on paper, you may have the same reaction he did when finding out what it actually does.

"That's it?" he asked.

The Scribe pen was designed to record audio within a 16-foot range with just a two-second press on the top of the pen. It transcribes the audio, provides summaries, answers questions about the conversations, and showcases the highlights or action items.

It's a perfectly capable product. When I used it to document the process of booking flights with my boyfriend, the transcript was accurate, even though the cafe was loud and I kept shifting the pen away from us to see if it could still hear our conversation. The summary graphic was so detailed that it remembered bits I hadn't even remembered, such as where our layover was, and presented it all in a digestible format. The more in-depth summary was accurate as well.

But it'll cost you. The Flowtica Scribe retails for $159, or $209 for the Scribe and charging case, which offers up to one full week of battery life. There is also a basic AI plan that includes 300 AI minutes per month for all advanced features, with tiered plans for additional AI access at $10 or $20 per month.

This highlights one of the biggest pitfalls of this category: the product comes at a surprisingly high cost for what it is because the hardware itself includes complex components, such as tiny processors and microphones. Plus, users typically have to shell out extra money for a subscription.

Last week, Plaud unveiled its Plaud One AI-powered wireless earbuds, which record and transcribe audio, including phone calls, online meetings, and in-person conversations. The Plaud Agent is meant to "help turn conversations into useful output."

The Plaud charging case includes an eSIM that has a wireless connection to bypass a phone. Again, I can't think of a situation where I would have earbuds and not my phone or laptop. And it costs $249, a price tag that rivals high-end ear buds like the AirPods Pro.

Our Deeper View

My concern is that this entire category of AI devices is watering down and muddying the definition of an AI product. Ultimately, AI is behind the scenes powering features small and large on most tech products. For instance, washing machines have used load-sensing and fabric-detection algorithms since the 2000s. By today's standards, would this make those AI products? I've been a fan of the AI hardware category since smart glasses were the only entrant. Smart glasses give AI context about the world around you, so you're not stuck feeding it that context yourself. Audio products technically do this, but audio is already easy to feed AI without extra hardware. Visual context is a harder problem. Short of holding up your phone in your line of sight at all times, there's no easy way to capture it. Solving this removes a genuine point of friction and lets you tap into AI in a far deeper way. That's the kind of thing we need out of AI devices: something you can't easily do with your phone.

Google has introduced Expert Intelligence, which is less flashy than its name suggests but still full of promise.

Gemini Notebook, previously known as NotebookLM, now lets you load ebooks in addition to the resources you could already add, including notes, PDFs, slides, web pages, and more. Once in the notebook, you can interact with it as you normally would, asking questions about it or doing the other things you can do in Gemini Notebook, such as making flashcards, building slide decks and infographics, and creating podcast-like audio overviews of the content.

At first, when I heard about the feature, I was really excited because it seems extremely valuable for readers. Often, when reading a book, you want to reference a particular part later or learn more about what you're reading. Being able to chat with Gemini Notebook about a book's contents would make either task easier and could even be a useful tool for improving reading comprehension.

However, there is a major caveat: You can only insert books from the Google Play Store that you have previously purchased.

The Google Play Store limitation restricts how many users can use ebooks to enrich their "notebooks" or information repertoires, since most ebook readers buy their books on platforms like Amazon Kindle. People are already expressing on X their desire for Kindle and Apple Books integrations, or even for academic journals to be referenced, such as MIT Press Open Access.

Yet that doesn't mean the feature should be discounted, as companies often first integrate features with internal products at launch and then, due to popular demand or after seeing how the trial run went, expand to other sources. Also, to help bridge that gap, Google is buying one book per person in the US while supplies last.

To see if a book you plan to buy is eligible for Expert Intelligence, visit Google Play Books, where a Gemini Notebook badge will be listed when you click the "Tools" badge on a book’s detail page.

Our Deeper View

In the era of AI slop and user distrust, it is more important than ever for AI companies to make it clear where their sources come from and ease concerns about AI hallucinations and accuracy. For that reason, since NotebookLM first came onto the scene, people have been excited about the tool because it lets users reference only their own sources and notes rather than scraping information from the internet, which can often be inaccurate. The Expert Intelligence feature builds on that by integrating ebooks, though it will need broader reach across more ebook publishers and platforms to be useful.

Nvidia may be taking one step closer to dominating the open-source AI ecosystem.

The tech giant has agreed to buy Hugging Face, the dominant repository of open-source AI models, for $12.9 billion, according to a report by The Information, citing a source familiar with the agreement. This is a strategic move that allows Nvidia to diversify its portfolio from just chipsets and take ownership of more of the AI stack.

This would give Nvidia ownership over the "front door for open AI innovation," Ashish Nadkarni, Group VP of enterprise infrastructure at IDC, told The Deep View. "Owning that front door gives NVIDIA a major position in the mindshare of today's AI development personas."

The deal comes as Nvidia faces mounting pressure from the very labs that rely on its chips. Google, OpenAI, and Anthropic are each moving toward custom or co-designed AI silicon, a shift that threatens Nvidia's near-monopoly on AI hardware. Worse for Nvidia, these labs bring something it can't easily replicate: deep, firsthand knowledge of the models the chips are meant to run, letting them tune hardware for performance in ways a general-purpose chipmaker can't match.

Hugging Face could eventually become a non-chip source of revenue for Nvidia, which could spur more growth if chip revenue eventually stops skyrocketing. A small caveat, however, is that, according to the report, Hugging Face has only generated $150 million in annualized revenue thus far. That makes Nvidia's payment about 80 times the startup's forward revenue, the report notes.

Revenue aside, the advantage of being both a model and chip maker is a key reason Nvidia has invested heavily in open-source models. Nvidia's commitment has been reflected not only in signing an open letter supporting the ecosystem, but also in internal investment in and development of open-source models, such as its Nemotron models. Of course, the Hugging Face acquisition moves that forward.

Additionally, there's a significant robotics angle: Nvidia already has a strong position with its Jetson, Isaac, and Omniverse solutions, and Hugging Face's own robotics efforts give Nvidia another point of leverage as physical AI becomes the next frontier. And just today, Hugging Face unveiled Microduck, a $399 open-source robot that learns new tricks via reinforcement learning. Even its looks resemble the Nvidia "Blue" research robot, inspired by Star Wars' BD-1, shown at demos, live events, and Disney.

Still, as Nadkarni highlights, the acquisition would also mean there would be less choice in the industry.

"To the extent that others like Intel, Apple, and AMD make similar plays, there will be fewer “ecosystem-neutral” platforms in the market," said Nadkarni. "Note how Microsoft has gradually made GitHub more like an extension of its own strategy (and for its own benefit)."

Our Deeper View

This is a landmark move for Nvidia, as it solidifies its efforts to be not just a chip company, but one that aims to own multiple layers of the AI stack at once. Hugging Face has become something like the GitHub of AI: serving as the industry's central hub for sharing models, datasets, and applications. Owning that layer, on top of the hardware layer Nvidia already dominates, deepens the moat. Already, nearly every open model runs on Nvidia GPUs, and buying the platform where those models live removes any ambiguity. It's also a defensive play: as OpenAI, Anthropic, Google, and Amazon push to build their own custom chips, controlling the distribution layer keeps developers tied to Nvidia's ecosystem regardless of who's building the silicon.

One of the most talked-about Google AI features from the Pixel 11 launch was Rambler, its highly effective speech-to-text feature. Now Google is spreading the tech that powers it.

On Wednesday, Google introduced Gemini 3.5 Transcribe, which the company describes as its most precise speech-to-text model yet, and capable of converting raw audio into accurate, polished text. Google shared that this is the model that powers Rambler on Android and the Gemini app on macOS, and it invites developers to build similar capabilities with the model.

Moreover, Gemini 3.5 Transcribe is available across two separate APIs: gemini-3.5-transcribe-live, which is focused on continuous bidirectional real-time interactions with sub-second latency and gemini-3.5-transcribe, which is focused on pre-recorded audio processing and transcribes recorded audio, meetings, and call logs with speaker attribution and word-level timestamps

What makes the model and the Rambler experience stand out is how accurately it can capture your ideas, even when they aren't presented linearly. For instance, during a correction, when one says "actually never mind how about…," it should not account for that; same with "umhs" and "ahs." Other features include:

• The model can delegate complex tasks to other Gemini models via function calls
• It recognizes jargon and unique spellings unique to the speaker
• It automatically transcribes over 85 languages

This progress is reflected in benchmarks, with Gemini 3.5 Transcribe outperforming its previous transcription model, Chip 3, by 70% on time to final transcription as measured by the Artificial Analysis benchmarks and on the FLEURS benchmark, which measures multilingual performance.

Anyone can experience the model today on Android via Gboard's new Rambler feature, in the Gemini app on macOS, and coming soon to Chrome. Developers can access it on Google Antigravity in Google AI Studio.

Our Deeper View

The most interesting part of this experience is that it will enable other developers to create experiences that perform as well as Rambler or even outperform it with post-training. The biggest competitor to Rambler at the moment is Wispr Flow, which has been dominating the market for months by offering the same thing to users across both Apple and Android devices, but it;s not nearly as seamless or integrated on mobile. With developers able to build similar experiences, Wispr Flow is going to face a flurry of competition, which has me wondering whether it will even be able to retain its place in the market. But the Google model is also exciting because, with so many more developers working on the same foundation, it should unlock even more features that make the model stronger. This is something to look forward to because it already works better than everything else other than Wispr Flow. In our recent podcast episode, we said that Rambler is so good that every other smartphone is likely to have a similar feature within the next two years. It feels like Google just made that a lot easier.