Skip to main content

Fast Company Technology

5 smart home devices that save you money (and 3 that are a waste)

2 weeks 2 days ago

The secret to smart home technology isn’t making your house feel like a sci-fi movie—it’s focusing strictly on your return on investment.

A handful of smart devices will pay for themselves within a year or two through lower utility bills and prevented disasters. Others, however, are expensive novelties that will never recoup their cost.

Here are five smart home upgrades that deliver genuine financial savings, along with three overengineered gadgets you should skip.

Smart devices that actually save money Smart thermostats

Heating and cooling account for over half of the average home’s energy consumption. Traditional programmable thermostats fail because nobody wants to manually reprogram a clunky digital interface every time their schedule changes.

[Photo: Ecobee]

Smart thermostats like Nest and Ecobee use geofencing and motion sensors to know when you’ve left the house, automatically easing back the HVAC. Most households save around 10% on heating and 15% on cooling.

As a bonus, many electric utility companies offer rebates just for installing one.

Smart water leak sensors

An unnoticed slow drip behind a washing machine or under a water heater can ruin floorboards, rot drywall, and breed mold long before you spot water pooling in the hallway. The average residential water damage claim costs more than $10,000.

A three-pack of basic Wi-Fi leak sensors costs less than $50. Tucking them under sinks, behind toilets, and next to your water heater ensures you get an instant push notification on your phone the second a drop of water hits the sensor contacts.

Plus, some home insurance carriers offer discounts on your annual policy premium if you install connected leak detection hardware.

Smart plugs with energy monitoring

Even when turned off, big-screen TVs, game consoles, audio receivers, and desktop computers continuously pull “phantom power”—or “vampire draw”—from wall outlets 24 hours a day.

Smart plugs featuring live energy monitoring show you exactly how many kilowatt-hours an outlet is consuming in real time. You can set auto-cut routines that completely sever wall power to media centers or home office desks overnight, saving you some cash in idle electricity costs.

Smart sprinkler controllers

Traditional irrigation timers are notoriously dumb. They fire off lawn sprinklers at 5:00 a.m. regardless of whether it rained an inch at midnight or a downpour is forecast for two hours later.

[Photo: Rachio]

Connected irrigation controllers such as Rachio sync directly with hyperlocal weather data to skip watering cycles automatically when rain, wind, or freezing temperatures occur. Homeowners with large lawns routinely cut their outdoor water usage by up to half during spring and summer months.

Smart light switches (not bulbs)

Replacing every light bulb in your house with a $15 smart bulb gets expensive fast, and if someone flips the physical wall switch off, the bulb dies anyway. Smart in-wall switches replace the wall controls directly.

Installing smart switches or motion-sensing switches in high-traffic zones (hallways, basements, kids’ bedrooms) ensures lights turn off automatically after periods of inactivity. Pairing them with LED fixtures keeps lighting costs down to pennies per month.

3 smart devices that are a financial waste Smart refrigerators with touchscreens

Paying an extra $1,000 to $2,000 for a giant tablet built into your refrigerator door is one of the worst tech investments you can make.

The internal computer and display will stop receiving software updates in 5 years, while the rest of the fridge is supposed to last 15. Furthermore, the internal cameras meant to “show you what’s inside while at the grocery store” are routinely obscured by milk cartons and leftovers. You’re paying a huge premium for a screen that will inevitably lag long before the compressor breaks down.

Motorized smart blinds

Motorized window shades that open and close automatically via app control sound luxurious, but the math behind them doesn’t add up.

Outfitting a typical home with custom motorized blinds costs between $2,000 and $5,000. While keeping shades drawn during hot summer afternoons does reduce solar heat gain slightly, the energy savings amounts to maybe $20 or $30 a year. It would take more than a century of energy savings to recoup the initial hardware investment.

Connected appliance dongles and ‘smart’ small gadgets

From Wi-Fi-enabled trash cans that order garbage bags to smart soap dispensers and connected toaster ovens, the consumer market is flooded with microappliances that add app connectivity just to raise retail prices.

These devices solve nonexistent problems while introducing security risks and unnecessary app clutter. They offer zero energy efficiency gains and carry higher repair failure rates than their simple, mechanical counterparts.

I auditioned 12 AI agents to be my chief of staff. Here’s how they compared

2 weeks 2 days ago

This article is republished with permission from Wonder Tools, a newsletter that helps you discover the most useful sites and apps.

I’ve never had a secretary before. Or a chief of staff. With these AI agents, I suddenly feel like I have both. This week, among other things, they:

  • Researched and booked Amtrak train tickets for me
  • Briefed me about important overnight emails and upcoming meetings
  • Surfaced an important contract I had overlooked
  • Turned my bookshelf into a visual reading list

Unlike older chatbots, these agents can take action proactively without waiting for a prompt. They can go online and research or buy things, make reservations, or fill in forms. They can look through various documents, transcripts, and emails to help you finish tedious projects.

Agents are the biggest step forward for AI this year. I’m testing a dozen to compare their reliability, features, and design: Thine, Instinct, ChatGPT’s Dots, Town, Hark, Muse, Poke, Ambi, Sentience, Extra, Slashy, and Claude Duet.

My early favorites: Dots, Slashy, and Hark
  • Dots has helped me spot calendar conflicts and find and prepare materials I need for work. I use it several times a day.
  • Slashy has changed how I manage email. It’s a strong alternative to Superhuman, the email app I’ve used for years. Slashy puts an AI agent inside my inbox.
  • Hark has a standout design. It’s been reliable for shopping and reservations.
Other observations
  • Business agents will be a separate category. Town, for example, is easy to use and well designed, but its features are tuned to small businesses and midmarket companies.
  • These are early impressions. I’ll share more as the tools develop. I’m working on a post about Slashy and how to use AI to help with email overload. The inbox is where I see early agents being most helpful.
One is enough

Start with one agent that does what you need. I’m testing a bunch for now, but ultimately I’ll prioritize just a few with distinct strengths.

What they cost

Some agents, like Instinct, Hark, and Muse, start out free. Others require a paid subscription.

  • Dots is so far available only on ChatGPT’s Pro plan, which starts at $100/month.
  • Town starts free, but the Pro plan is $49/month.
  • Hark starts free. Upgrade for $20/month for more AI usage or $100/month for full power.
  • Thine is free to start. It’s $15/month or $120/year for extra capabilities.
  • Pricing for other agents is still taking shape. You can test many for free. Before connecting your accounts, check on ongoing costs and how each service uses your data.
Connect your email, calendar, and files

Thine and Dots became much more useful as soon as I connected them to my email, calendar, documents, and Granola meeting transcripts. Now they have a broad view of my projects for context, as well as a detailed understanding of my emails and tasks. Drawing on meeting transcripts, they even understand what I’ve promised to do.

2 reasons to try an AI agent
  • They save time, handling pesky tasks that used to take hours. Examples:
    1. Hark coordinated a tricky travel itinerary and handled the booking on Amtrak’s clunky site.
    2. Dots pulled a batch of related emails into a doc, a task that would otherwise have taken me 45 minutes.
  • They prevent headaches. They don’t wait for me to send a prompt. Numerous times they’ve helped me avoid costly errors. Examples:
    1. Dots found an appointment mentioned in a buried email and alerted me that it was missing from my calendar.
    2. Thine reminded me to send an important email I had forgotten.
5 differences between agents and chatbots
  1. They’re easier to get started with. Claude Code, Codex, and tools like Hermes Agent are at least as powerful as most agents I’m describing here, but they can seem intimidating. Setting them up is a bit like learning to use a manual camera. You have to figure out which connections to enable, which plugins to add, and which skills to install. Agents like Thine, Hark, and Town are more like point-and-shoot cameras. They guide you through setup and connections to your calendar, email, Notion, etc., with explanations in plain language.
  2. You text them like a person. Many agents, like Instinct, Poke, and Slashy, work in SMS, iMessage, or WhatsApp. So you don’t have to open an app or browser tab. It’s like messaging a colleague instead of walking over to the help desk.
  3. They reach out to you. Chatbots wait for your prompt. You might go days without remembering to use one. Agents send morning briefings, flag missing details on your calendar, and nudge you about unsent drafts. It’s the difference between a chief of staff who knocks on your door versus a reference librarian waiting at a desk.
  4. Flexible models. Most agents can switch among models from different AI providers, like travel agents who can book any airline. ChatGPT, Gemini, or Claude rely primarily on their own models.
  5. You have one ongoing conversation, not hundreds of chats. I like how Thine and Dots keep a single long-running thread. Both remember our past conversations reliably. Chatbots I use scatter my history across lots of threads, which can make older exchanges hard to find.
The lines are blurring

For now the differences are noticeable, but major AI companies are adding similar capabilities. OpenAI’s Dots is already an excellent, if expensive, agent, and Claude Duet arrived on October 23.

I avoided the first generation of AI agents, like OpenClaw, because they seemed complex and risky. But several agents I’m testing are easy to use, and free to try.

What I hand off, and what I don’t

My goal isn’t to outsource my thinking or writing. I enjoy both. I want agents to handle background tasks like filing, organizing, and checking email for important messages. Or filling in annoying online forms. That leaves me more time to concentrate, with less worry that I might be missing an important message or task.

4 ways my AI chief of staff helps 1. Tame my inbox

I now get three times as many emails as I did three years ago. Many are AI-generated marketing or PR messages. With that email slop strewn across my three addresses, it’s easy to miss an important message among more than 1,000 emails a week. AI agents help me:

  • Summarize important new messages. At 8 a.m. my agents text me a summary of important emails from the previous evening and overnight. Instead of starting the day by scanning through hundreds of messages, I can just act on five important ones.
  • Handle irrelevant email for me. I respond myself to important and personal messages. But I don’t have the four hours it would take to reply to the hundreds of unsolicited, irrelevant emails I get every day. My agent can sort and archive those, or, in some cases, suggest quick “no, thank you” notes I can edit.
2. Plan my day

Mapping out my day at 8:30 a.m. means fewer decisions later, when I’m less focused, about what I should be doing.

My agents can look at my calendar, email, project materials, task list, and notes to suggest a detailed plan based on my preferences. They know I need breaks for stretching and meditation, a lunchtime walk, and time to call Mom. They use data from Rize, my time-tracking app, to estimate how long tasks usually take me.

During my Friday review, they help me reflect on next week’s priorities, tasks that need follow-up, and long-term projects to start. I head into the weekend feeling less stressed.

3. Research tricky decisions

In a day, I might have five to 10 questions an AI butler can help me research. I recently needed to turn 11 lengthy Google Docs into a printed book. I had to compare file conversion tools, check prices and reviews for book-printing services, and understand the technical steps.

That research would once have taken me an hour or more. This time I asked an agent to handle it and went back to preparing to teach. About five minutes later, my agent delivered an excellent recommendation in the Google Doc I’d requested.

I don’t like shopping. But I like knowing I’m getting a well-reviewed product or service at a good price. Hark and Instinct can compare options, recommend what to buy, and set up the purchase for me to confirm. Caveat: Check how free shopping agents make money, and whether their recommendations include paid placements.

4. Organize files

Connect an AI assistant to your files and work apps, like Google Drive, Slack, and Notion. Then ask it to move documents between folders. You can ask it to summarize document changes or ticket activity, or to create a new dashboard.

The hardest part is deciding what to delegate. I often ask my agents to suggest useful tasks they can handle. I tell them what tasks I find most annoying, what I enjoy doing myself, and what I can’t risk messing up.

Caveats
  • A new monthly bill. If you already pay for ChatGPT, Gemini, or Claude, adding another AI bill may seem extravagant. But for what will likely end up costing about $5 a week, you get a 24/7 assistant. If that saves you time and stress, it may feel worth it. As competition heats up, your main AI subscription may soon include an agent. For now, it’s like streaming. Netflix alone might be enough for some, but others happily pay for Hulu, Apple TV, and Disney+ too.
  • Expect periodic roadblocks. The services are still new. One told me it could send me a text, but not an email. Another said it could open docs from only one of my Google accounts. These annoyances can stall an otherwise useful task. Some problems have quick fixes. Others require workarounds while these services develop. I remember having to use two separate computers when dial-up Internet meant waiting minutes for each Web page to load.
  • Connecting your accounts can be a nuisance. Assistants need permission to access your accounts. That’s important for protecting your security. But it can create headaches when you have multiple Gmail accounts, extra sign-in steps, or workplace restrictions. I’m glad security protections prevent random people from logging into my accounts. But expect some setup friction.
  • Some won’t survive. For now, the money is flowing freely. Town raised $108 million this week at a $1 billion valuation and Instinct raised $1 billion recently. As individuals and businesses choose their agents, subscription fatigue may limit how many they’re willing to pay for.

The bottom line: If you haven’t yet experimented with an AI agent, start a free trial. Assign a few research, summarization, and shopping tasks. Judge it by the usefulness of its response, the time it saves, and how easy it is to use. Leave a comment to share your impressions.

This article is republished with permission from Wonder Tools, a newsletter that helps you discover the most useful sites and apps.

Why driverless trucks won’t be taking over highways just yet

2 weeks 3 days ago

Driverless trucks with artificial intelligence on board could dramatically change the landscape of freight shipping. A few companies have begun testing early models on some limited routes in Texas and California. Simulations of automated fleets show the potential benefits of the technology: The vehicles could improve safety, enhance fuel economy, and lessen transit times.

Truck makers and operators still have to solve some major issues, however, including how automated trucks will operate in variable weather and road conditions, and whether drivers should still be inside big rigs.

I am an engineer who studies human factors and transportation safety, and in my view, it is likely that automated trucks and human drivers will be working together for quite some time.

Benefits of automation

Fully automated trucks, which can drive on an interstate highway in good weather without human intervention, use a combination of AI, machine learning, computer vision, pattern recognition, and adaptive decision-making. Torc, one company developing these vehicles, uses AI to combine data from multiple sensors and generate models that predict truck performance.

Data taken from modern big rigs that are equipped with certain technology such as automatic emergency braking and lane-keeping assistance systems indicate that fully automated, driverless trucks could reduce the number and severity of crashes. They could also save fuel and therefore cost by optimizing acceleration and braking, and by running multiple vehicles in a close-knit convoy or platoon. Driverless trucks could reduce parking needs as well as emissions from idling because they could run almost 24 hours a day and would not need to idle when they do park, which is common for regular rigs at truck stops.

Finally, automated trucks could help ease the ongoing driver shortage. Ironically, automation could make driving less stressful and less physically demanding, improving job satisfaction among drivers.

Weather and work zones are hard to navigate

The main reason automated trucks are not common today is that they have difficulty operating in variable weather, under changing roadway conditions, and alongside unpredictable drivers.

Automated trucks require sensors, radars, and cameras that gather information about the environment around them to make decisions. However, snow and rain remain a problem for all automated vehicles, including cars, because they prevent the technology from “seeing” the roadway. Driverless taxi company Waymo, for example, operates driverless cars only in areas that get no snow.

Furthermore, any stretch of road where there may not be a predictable, clear surface and prevalent lines is difficult to handle. Imagine a construction zone with orange cones, narrow lanes, and missing lane lines. Driverless car companies still try to avoid these areas. And, oddly, some research shows that automobile drivers tend to crash into driverless cars at a higher rate than they do cars piloted by people.

Given those challenges, companies are testing automated trucks on roads that lack complications.

For example, autonomous truck maker Aurora is operating large freight trucks on I-45, a relatively flat and straight highway between Dallas and Houston. Rival Kodiak’s robotrucks are delivering sand to industrial sites in Texas over roads few people use. Few if any automated trucks are being tested in chaotic urban environments, given the pedestrians, bicyclists, traffic lights, and elements of city traffic.

Driverless trucks are operating on some roads in Texas.

Some companies that were developing automated trucks have scaled back ambitions, changed direction, or exited the market.

In 2023, Embark, TuSimple, and Locomotion closed down. Waymo began to focus solely on cars. Other companies narrowed their focus to trucks that would operate only along a few dedicated routes between two large cities or manufacturing sites.

Drivers and trucks that work together

In my view, for years to come, automated trucks with AI-based controls will still require a driver.

The best technology will support a symbiosis between human and machine, where the person and the truck each do what they do best. A human would likely drive the truck across the “first mile”—the uncertain and crowded warehouse grounds and local streets to a highway. The driver would then stop using the steering wheel, gas pedal, and brake and let automation take over. The human would take back control during the “last mile” to the truck’s final destination.

Research my colleagues and I conducted shows that human-machine interfaces that communicate between drivers and autonomous vehicles will be key. For example, a screen or head-up display inside the cab would tell the driver how the technology is doing. Verbal messages or a vibrating seat could signal that the automation cannot handle a developing situation, and the driver would take over.

Developers will have to be aware of possible biases. So-called automation bias could lead to a driver who trusts and relies on the truck’s decision-making so much that they defer to it. Algorithmic bias could cause truck sensors to recognize certain groups of people on the road more than others, such as only detecting pedestrians of a certain size or shape.

The long-term vision is to have no driver, in part to lower costs. But driverless trucks do not seem to me to be practical on a large scale in the near future. It is more realistic to envision the driver and the automated truck as a team. Vehicle manufacturers, technology developers, regulators, and drivers will need to work together to realize the benefits.

Shannon Roberts is an associate professor of industrial engineering at UMass Amherst

This article is republished from The Conversation under a Creative Commons license. Read the original article.

How HomeView foreshadows Apple’s post-app, Siri-first future

2 weeks 4 days ago

As the iPhone grew into an unprecedented success, Apple was marching its users toward millions of downloads to the drumbeat of “There’s an app for that.” Meanwhile, Amazon and Google were flooding homes with waves of TV platforms, smart speakers, and cameras. These products and their underlying connectivity standards formed a deep foundation to support a new generation of AI-driven services far surpassing those of Google Assistant and the original Alexa.

At its home product launch on October 13, Apple updated its Apple TV and HomePod mini speakers. But all eyes were on HomeView. The touchscreen family hub is spinning and tilting the company toward a new direction led by a reinvented Siri and away from a dedicated app-first ecosystem.

If the iPad was originally dismissed as a large iPod Touch, the HomeView—with its 1:1 aspect ratio and widget-driven display—may at first evoke a large Apple Watch. But not for long. The device can be mounted on a wall or tilted on a domed aluminum speaker base that has the metallic texture of a Mac mini as opposed to the cozier cloth stylings of a HomePod mini. And if the frenetic 3-minute family morning tableau Apple produced to showcase the device proves accurate, we can expect it to serve up a whirlwind of multimodal interactions as opposed to the tap-and-takeaway interactions of Apple’s wrist computer.

On the other hand, HomeView’s platform philosophy borrows from another Apple experience: CarPlay, which presents apps installed on the connected iPhone on automakers’ dashboard displays. While HomeView is an independent device, it still has strong ties to the iPhone, which provides its personalization and app engine.

Amazon and Google smart screens have long featured cameras and personalized their experiences for recognized individuals. But Apple has enabled much richer personalization on the fly by using iPhone proximity to determine who’s using HomeView. It’s a best-of-both-worlds approach that Google should absolutely adopt with Android should it revive its Nest Home displays. The approach would also benefit both the Apple TV box and service, the latter of which trails Netflix, Disney+, Peacock, and others in supporting viewer profiles.

No App Store, but all Apple

Though HomeView offers far more of a standalone experience than CarPlay, it doesn’t have its own app store. Nor does it introduce a “homeOS” (Apple says it’s powered by “HomeView software”) to complement Apple’s family of operating systems for its computers, phones, watches, and TV devices. It‘s Apple’s first “post-app” device, similar to the way the iPhone was its first major “post-PC” platform.

Of course, that doesn’t mean that apps have vanished, any more than PCs (including Macs) did in the post-PC era. But as it created a device that serves as a primary interface for smart homes, Apple—like Google and Amazon before it— has recognized that an app store is not the ideal approach.

Developers can indirectly introduce apps onto HomeView by including support for the device in their iPhone apps, the same way CarPlay works. In fact, third-party iPhone audio and video apps that support CarPlay, such as Spotify and Audible, should work out of the gate on HomeView.

However, this approach can lead to relevant apps not getting noticed by consumers, if they’re not as compelling on the iPhone. Conversely, less relevant HomeView apps could also show up on the device simply because they’re companion components of an iPhone app.

In any case, while the HomeView may not have as rich an app ecosystem as the iPad, it’s still relatively app-centric compared to its Amazon and Google counterparts. Moreover, customers will appreciate that there isn’t an ad in sight, in contrast to Amazon’s Echo Show.

Home hubs such as HomeView pose a unique privacy challenge because they are far more likely to capture information that some family members might want to keep private from other family members than, say, a streaming device like Apple TV. As major ecosystem vendors feed far more intimate knowledge about us into AI than was ever captured for advertising personalization, the stakes for privacy will rise even further.

HomeView provides an exceptionally good platform for Apple to promote its commitment to privacy protection. In addition to the iPhone-driven personalization, communications are protected by Apple’s Private Cloud Compute. The device’s rollout also marked the debut of cameras and thermostats from LG and Schneider that feature high-end design and adhere to Apple’s privacy standards.

Despite the appeal of the privacy pitch, it’s unlikely that HomeView will be much of a beachhead into homes that aren’t already deep into Apple’s ecosystem. It’s enmeshed in a web of Apple services and technologies beyond Siri, including Apple’s cloud, video and music services, FaceTime, Home, Safari, Handoff, and even Find My. (Despite the focus on touchscreen interactions, there’s no support for Apple Arcade or iOS games in general.)

Striving with Siri

Like other smart screens, HomeView has a rich touch interface. However, as Apple’s first post-app device, it is also the company’s first Siri-first device. The past few iPhone launches have leaned heavily into spotlighting Apple Intelligence and showing how its revitalized agent is creating and improving new experiences, but smart screens are a new twist, and are often accessed from across the room. The new Siri’s more natural conversation and far greater knowledge and skills make HomeView a much more compelling proposition, one that stands to improve more rapidly because it learns consistently from iPhone usage.

This year has seen Apple jump into categories where competitors had a longstanding presence. Samsung introduced the first Galaxy Z Fold in 2019 and Amazon’s Echo Show debuted two years before that. Functionally, Siri closes the gap with devices that only recently got access to Gemini Home and Alexa+. And true to form, Apple has leveraged late market entry to shake up some category assumptions.

HomeView’s mix of closeup and distant functionality has raised the bar for the incumbent devices from Amazon and Google. And its departure from Apple’s standard practice of dedicating an operating system and app store to each major device category could presage a new era for the company’s devices.

Otter.ai is moving beyond transcription—and deeper into your workday

2 weeks 4 days ago

Eight years ago, Otter.ai’s pitch for its transcription app was pretty straightforward: The Mountain View, California-based startup would turn audio into readable text in real time. And it would do it for free. But in 2026, the company is now competing against a chorus of competitors that all leverage AI to provide real-time transcription. Otter’s answer to that challenge amounts to changing the channel—to go beyond transcribing to extracting insights from all that banter.

“As our CEO Sam [Liang] has shared a number of times, a transcription was never the end goal,” says Otter’s new CMO Alex Gay. “The opportunity for us really is about what can you do when you capture all of that conversation across all the contexts that organizations need.”

As in, building on that to make meetings less of a time-suck by going beyond Otter’s existing ability to summarize what people said (a feature many competitors can do) to suggest what you should say next. 

Gay argues that meetings are often inefficient for predictable reasons: Participants arrive unprepared, the right people aren’t in the room, or agreed-upon follow-up tasks never get done. The result is what Gay calls “this loop of having to have more meetings to just improve the productivity of the meeting.”

Otter hopes to break that cycle with three new features introduced Tuesday, all built around what Liang calls “the whole meeting life cycle” during a recent Zoom call.

All of them reward organization-wide use of Otter, which offers a free Basic tier with usage limits as well as three paid tiers: Pro, $99.96 per user per year; Business, $239.88 per user per year; and Enterprise, with pricing based on such factors as the size and scale of a deployment. Otter says it has more than 40 million users but doesn’t break down how many pay.

First, Otter will automatically compile a “Pre-Meeting Brief” based on transcriptions of prior meetings as well as whatever other apps and tools an office connects to Otter via the platform’s array of integrations.

Second, Otter will offer “Live Assist” help during Zoom, Microsoft Teams, and Google Meet calls based on the context of the conversation. Liang describes it as a “sales-call sidekick” that’s “listening to the customer. It has all the background on the customer. It’s trained on thousands of high-level sales calls in the past.” (While other participants with an active Live Assist agent will see an Otter notetaker in the meeting window, its full capabilities are not disclosed.)

Afterward, Otter switches to “Work Mode” to put together a recap and help handle follow-up tasks, including drafting emails, Liang says.

The CEO’s description for the app that he doesn’t want people to think of as a transcription app: a “conversation knowledge engine.”

Otter also plans to expand language support beyond the current limited selection of English, French, German, Spanish, and simplified Chinese while further improving its understanding of accents. (Gay, who speaks with a Scottish brogue, admitted that Otter’s accuracy can drop “when I get particularly Scottish or start to use colloquialisms.”)

And the company plans to step up its marketing for all of these features with a new $320 million investment from General Catalyst. Gay says expanding Otter’s adoption in organizations beyond individual- or team-sized beachheads is high on its to-do list.

“We think we have to demonstrate the impact that we can have across two or three functional areas, and then show how that could be improved with context access across their organization,” he says.

Privacy concerns

AI notetakers on calls haven’t always been welcome, with some security agents advising against using them at all. Otter itself is defending a class action lawsuit filed last year alleging that its software “deceptively and surreptitiously” captured private conversations.

State privacy laws regarding recording conversations vary widely. California, for example, requires the consent of all parties in a conversation, while Virginia only needs the consent of one party.

Gay acknowledged those concerns but maintained that Otter doesn’t repurpose customers’ conversations for other uses. He says the company’s model training is limited to speaker identification, rather than mining the contents of conversations as some other generative AI tools might. (I should note that I used an AI-transcription app for our conversation, but one that works offline: Google’s Recorder app for Pixel phones.)   

Gay also pointed to Otter’s use of encryption both at rest and in transit, arguing that customers can be confident their data remains their own. Still, the CMO acknowledges that Otter has more work to do to satisfy some security-conscious customers. He says the company already serves regulated industries, including finance, but needs to develop additional capabilities for organizations with zero-trust IT policies. “There’s still new capabilities that we need to build.”

What else is listening?

In a world where a smartwatch, smart glasses, or even AI jewelry could always be listening by default, Otter’s app-based approach could seem a little old-fashioned.

Asked about whether Live Assist could be something that runs on a pair of connected glasses, Liang waved off the idea. “It’s an interesting idea, but we’re not doing it right now,” he says, opining that camera-equipped smart glasses are “just too creepy.”

Gay raised a related question about always-on microphones: Do people begin to censor themselves when they know they’re being recorded? But he suspects that discomfort will fade as the technology becomes more commonplace.

“Eventually, it just becomes the norm and then people become relaxed,” Gay says. The larger question, he suggests, is who ultimately benefits from all those recorded conversations.

“Is it being used to the value of you, as the person that’s being recorded? Or do you feel like it’s just your words being taken to benefit someone else?”

Enterprise AI is waiting for its PageRank moment

2 weeks 4 days ago

Let’s try a little thought experiment in technology-fiction: what would have happened if Larry Page and Sergey Brin’s PageRank, back in 1998, had failed to attract its first investors? Google didn’t invent the search engine: there were plenty of them before, from AltaVista, Excite, Lycos or others who were counting words on pages and selling their results page to the highest bidder, to catalog or portal concepts such as Yahoo! or AOL, that attempted to capture the users’ attention as much as they could to be able to serve them advertising, the more the better. Imagine if these concepts would have gone the distance, and deprived us of the web that resulted from Google’s easy and efficient way of searching … 

What did Larry Page and Sergey Brin realize? They adapted and extended the citation index (or its consequence, the citation impact), a system that they knew worked for academic literature (remember, these guys were graduate students working on their Ph.D., and thus reading academic papers like there was no tomorrow!), and moved it into a new context: the web. They realized there was an underused signal, relationships among pages, and used it to generate a metric, weighting links according to the importance of the pages providing them, that not only allowed them to index the web, but also to define attention as a social construct: a page is important when other pages link to it with a certain term. 

What’s today’s underused signal in corporate AI? 

As I’ve said several times, corporate AI is completely different from individual AI: companies don’t need just answers to questions, what they need is persistent memory and state, feedback learning loops, embedded cybersecurity, logs that can be tracked and not manipulated, etc. 

So far, companies are building these required scaffolding using twentieth-century tools such as Python, and AI agents are proving to be smart enough as to escape from these containments and do all sort of bad things — not because they are intrinsically evil (they cannot be), but because the goals and the guardrails are badly defined, or due to excessive permissions, prompt injection, or weak isolation. These companies are trying to define these properties by repeatedly assembling application by application, rather than supplied coherently by the platform. 

However, companies are repeatedly making all sorts of decisions about suppliers, customers, inventory, pricing, service, etc. Those decisions are constantly generating evidence of what works and what not, provided that the company is smart enough to capture them. What if we could evaluate AI by whether the company learns and improves, beyond the quality of individual outputs? What if AI optimization becomes more important than having access to bigger models? 

The competition moves from model capabilities to system performance 

A company does not just need a very powerful AI model: it needs an architecture that coordinates capabilities toward its goals. Researchers writing on Berkeley’s AI research blog explicitly describe the shift toward compound AI systems, where system design can offer better returns than further model scaling. 

Other experiments, such as FrugalGPT or DSPy are specific precedents for making model selection subordinate to results: FrugalGPT demonstrated that model cascades can match the strongest individual model with up to 98% lower cost on its evaluated tasks, and DSPy showed that optimized pipelines using smaller models can compete with approaches using much larger proprietary models. 

What makes the next step consequential? The ability to learn across executions. If we manage to make each interaction inform subsequent decisions across the relevant population, we can achieve a substantial advantage. And world models add precisely this possibility: they can anticipate consequences before acting. Research about Dreamer successfully demonstrates this mechanism across diverse control tasks. And once you have that, you can use reinforcement learning or another optimization mechanism uses experience to progressively improve decisions. 

Accumulated organizational learning becomes the true competitive advantage 

While broadly available models available through commercial APIs or open-weight deployments can provide shared capabilities, company-specific experience can produce truly differentiated decisions, the ones that can really make the difference. Imagine, for instance, the renewal decisions you can make when you can learn from observed retention outcomes. 

However, bear in mind that simple accumulation does not automatically guarantee an advantage: you have to process the information properly, but not “with an AI”… with “your AI”. Make sure you put the training in the correct layer, the one you can administer and control. 

PageRank unlocked value in something the web was already producing: links. The next breakthrough in corporate AI could unlock value in something companies already produce every day: experience. If that experience becomes accumulated learning that the company controls and retains as models change, the competitive advantage starts to belong to the business itself. The company that stopped waiting for a smarter model might simply be the first to realize that its most valuable intelligence is the intelligence it learns to build — and keep. 

My proposed buying criterion is based on the following question: can the company replace its model provider while retaining its operational history, learned policies, objectives and demonstrated gains? Can your company export and reuse that learning when changing platform providers? Do you own your corporate intelligence, or do you rent the whole thing? When you stop working with “vanilla AI” and you start working with “your own AI”, you control the learning asset, without requiring ownership of every underlying model. 

And that’s when you can become a truly optimizable company. Basically, optimization, not a bigger model, is what you need. 

Apple’s ‘Write with Siri’ prompt was driving me crazy. Here’s how I shut it off

2 weeks 4 days ago

There’s a lot to love about iOS 27 and macOS 27, the new operating systems released for iPhone and Mac last month. The Liquid Glass design language now has a fluidity and polish that it lacked when introduced last year in iOS 26 and macOS 26.

But as much as I love the new operating systems, there’s one thing that has driven me crazy ever since I updated my devices: the “Write with Siri” tool, which appears as a popover floating toolbar on the Mac and a big button above the keyboard on iPhone.

The tool, part of the new Siri AI, helps users generate text nearly anywhere—in a document, email, or messaging app. Unfortunately, it’s there even if you don’t want to use it. (And Apple, I never do. I don’t need help from generative AI in deciding how I should respond to my mom’s text inquiring when I’ll be over for dinner next).

Its presence is not only pointless for many who don’t need machine help with their writing, but it’s distracting, especially on the Mac, where a stray mouse click can easily expand the popover when you are trying to edit text. Thankfully, there is a way you can disable Write with Siri—though Apple doesn’t make it obvious.

How to disable ‘Write with Siri’ on an iPhone or iPad

On the iPhone, the Write with Siri tool primarily takes the form of a big glaring button above the keyboard on iOS 27. It’s there by default, whether you want it or not. And frustratingly, if you do want to disable it, you can’t do so where you would naturally assume you could: in the Siri section of the Settings app.

But there is a place where you can. It’s hidden under the Content & Privacy Restrictions of the Screen Time settings in iOS 27. Here’s how to find those settings and shut Write with Siri down:

  1. Open the Settings app.
  2. Tap Screen Time.
  3. Under the Restrictions heading, tap Content & Privacy.
  4. Toggle the Content & Privacy Restrictions switch to on (green).
  5. Tap Siri.
  6. Tap Writing Assistance.
  7. Tap “Don’t Allow.”

Now, when your iPhone’s keyboard next appears, the big “Write with Siri” button sitting atop the keyboard should be gone, replaced by the default predictive text choices that have been a staple of iOS for years (and which are easier to disable, by going to Settings>General>Keyboard).

How to disable ‘Write with Siri’ on a Mac

On a Mac, things are a little less straightforward. Common sense says you should be able to follow the steps above to disable Write with Siri. But in my testing, even after turning off the Writing Assistance toggle (in the Siri content restrictions in macOS’s Screen Time settings), the annoying “Write with Siri” floating bubble still appears wherever you type in apps, like Pages. I suspect that this is an unresolved bug.

Thankfully, there’s another option for turning off Write with Siri on the Mac (though it is more restrictive). You can do this by disabling the new Siri AI in one of two ways.

The first is by going to the Siri settings in the Settings app:

  1. Open the Settings app.
  2. Click Siri.
  3. Click Turn Off Siri.
  4. Click the next Turn Off Siri button, too.

Alternately, you can disable the new Siri AI and default to the old, less robust version, which does not support the Write with Siri popover floating toolbar.

To revert to the old version of Siri: 

  1. Open the Settings app.
  2. Click Screen Time.
  3. Under the Restrictions heading, click Content & Privacy.
  4. Toggle the Content & Privacy Restrictions switch to on (blue).
  5. Click Siri.
  6. On the next screen, under “Allowed Siri Version,” select “Siri Classic” from the drop-down menu.
  7. Click Done.
Apple doesn’t make disabling ‘Write with Siri’ straightforward—and that’s probably by design

Big Tech has spent hundreds of billions of dollars on artificial intelligence development over the past five years—and there’s no sign that spending will be slowing anytime soon. It’s a huge bet, and in order for it to pay off, the tech giants need everyday users to adopt their AI tools en masse.

For many consumers, that adoption has felt forced upon them. Google search now surfaces AI answers before search results, Meta has shoved its AI assistant into nearly every app it makes, and Microsoft has integrated AI across Windows.

Apple, of course, has spent far less on AI buildout than its competitors have, but that hasn’t stopped it from jumping on the AI train. The headline feature of its new operating systems this year was Siri AI, and its newest hardware product, the HomeView, relies entirely on artificial intelligence.

That’s not to mention the fact that investors seem to be demanding Apple “catch up” to its other Big Tech competitors to become a major player in the space. So it’s no wonder that Apple, too, wants to put its AI offerings within as easy reach as possible—even if its users haven’t asked for it.

Increasing Siri AI usage is surely a goal the company has, so it’s not much of a surprise that Apple has made disabling its individual features less than straightforward, but now you know you at least can. I wonder what Siri would say about that?

This free creative app is like Internet Archive meets Pinterest

2 weeks 5 days ago

Do you remember, as a kid, going through a stack of old magazines and clipping out the sections that you liked? There was an art to it.

Whether you were in class or just making some decorations for your wall at home, it was important to find clips that expressed your personality—pictures you thought were interesting or cool or that in some way said something about you. And it was fun to find sentences to put out of context, or even to cut out individual letters to create your own custom messages.

But, much as some of us around here may still love them, physical magazines aren’t as omnipresent as they used to be. So how can you recreate that feeling today—and find new ways it can be useful or even just plain enjoyable in your adult life?

This tip originally appeared in the free Cool Tools newsletter from The Intelligence. Get the next issue in your inbox and get ready to discover all sorts of awesome tech treasures!

Your modern-day collage experience

​Clipart Studio​ is a website that simulates the experience of cutting up magazines to make a collage.

➜ You can page through real vintage magazines, virtually “cut” what you want, and put everything together in any way that suits your fancy.

⌚ You’ll need a mere minute or two to explore the site. Just open the website and get started.

In the left panel, you’ll find six random magazines, sourced from the Internet Archive​. You can browse by topic or click the little die icon to get six more random titles.

✅ Click a magazine, and you can start flipping pages—just the way you remember. When you see something you like, use the virtual knife to cut it out, then place it on your board. Keep doing this as long as you like.

Clipart Studio brings old-fashioned magazine collages into the digital age.

☝️ There’s a bit more here, too. You can search for images, sourced from Wikimedia Commons. Or you can drag an image over from your own computer, if you want to make sure your cat (or child, coworker, or whatever the case may be) is featured. 

With sources from the Internet Archive and Wikimedia Commons, you’ve got plenty of material to work with.

The possibilities are practically endless, whether you end up sharing your creation, using it for some social media purpose, or even just enjoying it yourself and not trying to turn it into anything more.

(Side note: The site was made by a designer named ​Sami Smith​, who has a knack for creating fun projects like this. Check out her ​version of Wikipedia running on an imaginary Windows XP desktop​ for another memorable retro-tech adventure. )

  • Clipart Studio works ​in any web browser​, on any device.
  • It’s free, and there’s no advertising, either.
  • Your collage is stored in your own local browser, meaning it’s entirely private (though you can optionally share it for a public gallery, if you want).

Treat yourself to all sorts of brain-boosting goodies like this with the free Cool Tools newsletter—starting with an instant introduction to an incredible audio app that’ll tune up your days in truly delightful ways.

Mozilla has a plan to rescue us from the zombie internet

2 weeks 5 days ago

There is an uneasy feeling these days when surfing the web that the internet has irrevocably changed. Log on to social media and you’re wading through a sea of influencers psy-opping you into buying stuff you don’t need. Scroll through LinkedIn and you can barely tell whether a post was written by a human or generated by Claude to rack up clicks and likes. Everywhere you look, the internet feels a little less human—and increasingly, it is. Recent studies suggest that at least half of the content posted online is AI-generated.

How, then, do we reclaim the internet as a space for genuine human connection and creativity? That’s partly the question the Mozilla Foundation, a 23-year-old accessibility-focused nonprofit behind the Firefox browser, is trying to answer with this year’s MozFest in Barcelona.

Held this week, the Mozilla Festival is an annual global gathering of developers, policymakers, and tech builders focused on building a more human future for technology. This year, the festival asked how to make the internet more human—or, in its words, how to “rewild” it. 

Fast Company spoke to Nabiha Syed, executive director of the Mozilla Foundation, about this year’s festival, its themes, and why she believes nothing about the internet’s future is inevitable.

This interview has been edited for length and clarity.

What does “rewilding” mean in the context of MozFest?

We can acknowledge that we live in a moment where the internet sometimes feels like a monoculture. There’s like six dudes in hoodies who are making decisions about what everybody should be consuming: the platforms we should be on, the things that shape the market. And what we know from nature is that monoculture works until it doesn’t—until there’s a collapse in an ecosystem. 

So much of the open-source ideology that is core to Mozilla is about creating the structures for things to be wild and unpredictable and to go off in different directions and then later cross-pollinate. Last year’s theme for MozFest was “unlearning”: “Hey, we need to pause. What got us here isn’t necessarily gonna be what gets us to the next place.” Wilding is about saying, “What would it look like if we just let things bloom? What if we had more people have more say over the way that we think the internet’s supposed to unfold and be?” 

How would you describe the wild internet? And what is illustrative of a monoculture internet to you?

I think about when I first learned how to code it was because I really wanted my MySpace to be amazing. I would view the source and make my page look like whatever I wanted it to look like. When I couldn’t quite get it that way, I’d go to GeoCities and make my own site in my own way. And there was freedom, interoperability, and contestability. 

Even in the blogosphere, the way that we spoke to one another felt like we were actually having conversations and not worrying about bot interjections. There was a feeling of possibility and not just predetermined ways of interacting, like, “You can comment in this way” or “You can engage in this way.” 

You got to create and build with the architecture as you wanted it. Every generation is always creative on any communications medium, but within the confines of a platform like Instagram. They can be innovative in how they use Finsta, or the way they shitpost, or the way they talk. But it’s still within the architecture that someone else has designed. 

What the early internet felt like to me, and to so many people who have nostalgia for that time, is that you could play with the rules of the game, and they were committed to interoperability, transparency, and being able to peer under the hood. 

What are the consequences of that lack of interoperability, or the confined nature of our current platforms?

To me, the big thing is that it creates what we call imagination deficit. If you think that innovation comes from what the platforms provide to you—and your job is to play within the sandbox as it is created for you—it starts to inhibit your understanding of what is possible. Everything’s gonna be an endless scroll. Everything starts to look the same. Everything starts to look like it’s Instagram but for X. It’s Uber but for Y. Things start to become cookie-cutter.

Over time that changes our understanding of the possibility, which is ultimately very bad for, you know, big concepts like our cognitive liberty, but also our ability to play online. 

I think what was so special about the early internet is the feeling that you could just make new things and people could stumble upon them. It wasn’t mediated by algorithmic pseudo-wisdom or nudging you towards commercial ends. We are creating for one another—not to sell a course or become an influencer, or that always-lurking commercial logic that’s on these platforms. Removing that logic, or at least creating other spaces where it doesn’t matter as much, is what’s so interesting about early internet times versus platform-dominated times.

Right. Well, I guess then the question is, where are those spaces online undominated by commercial incentives? 

There are two kinds of spaces. You can see it in “dark social” environments. You see it in the group chat, you see it in the Signal chat, you see it in these closed spaces where people feel like there can be a clear organizing principle of why you’re there. You’re not trying to show off for everybody. We’re here to actually communicate and create for one another. 

The second are [part of the] Dark Forest Operating System, some of the Yancey Strickler [cofounder of Kickstarter] stuff of re-creating slightly bigger groups. What’s interesting to me about DFOS is that that mentality feels a lot like kind of bulletin board-y vibes. It feels like we’re re-creating spaces that were very Usenet-y. But they’re there, and if you know, you know. You’re invited into them. And I think that’s useful.

I don’t think everything needs to be radically, radically, radically open in order to have meaning. That’s not how human beings organize themselves. But what I worry we’re losing is that if you only have nodes of dark social, or dark social types of principles, and no common, or not as many common places to all be together, then we endlessly silo. And then we lose sort of the moments and spaces of cross-pollination.

Do you see any open spaces that still have that ethos?

Rudy Fraser from Blacksky is one of our fellows. We’re actively celebrating him as a fellow and supporting his work as a fellow. I think there are experiments there, and we want to encourage them.

One of the reasons we have MozFest is that you want enough of those spaces to have moments of intersection and cross-pollination. I think we’re trying to figure that out. We’ve been to a place of radical openness of Twitter, where everything is out there. Then everyone is swinging to, like, the dark social, where it’s, like, the eight people in your chat.

And now, if there are community-run federated platforms, then the work is like, “Okay, it’s not super small, it’s medium-size. How do they intersect in a mosaic together? Where does that happen?” And that’s really a lot of the purpose around MozFest: to have people’s own communities intersect with other communities in a dynamic way.

Let’s talk about that a little bit. You mentioned Rudy from Blacksky as a Mozilla fellow. I think we should give readers a little bit of tone-setting about what it means to be a fellow at Mozilla, for people who might only know Mozilla as the cute fox browser. What programs establish these spaces and foster these connections?

Thank you for the invitation to share that: Mozilla Foundation’s core thesis at this moment is that imagination is not the domain of Silicon Valley. It belongs to the world, and that reflects the core initial promise of the internet. 

Our work is to build from that imagination. We know that people are imagining different technology futures that are actually pro-human agency, pro-community—but they don’t always get the resourcing and the support that they need. Our intervention’s into that: our fellowships.

Over the last few years, we’ve supported over 200 different fellows who share our one fundamental belief, which is that no part of technology is inevitable, and we get to decide. And that can be everything from spinning up OpenNews and talking about a new way of doing journalism, to having fellows like Camille François, who just started Robust Open Online Safety Tools (ROOST), to having fellows like Rudy to having fellows like Matisse, who’s working on open-source quantum.

There are a lot of different areas of technology that all benefit from the idea. We’re looking at cybersecurity. The internet is not what the internet was 25 years ago. It shows up in all these different ways.

What kinds of projects does Mozilla’s incubator fund?

Earlier this year, we incubated Mozilla Data Collective, where we invested $10 million into a double-sided data marketplace that has ethical, fair licensing as the core of it. For a lot of creatives, they’re not against AI wholesale. They’re just against having their content taken and not getting compensated for it, and used to enrich a few. So creating a data marketplace where people can actually decide the terms under which they license their data is a way of having infrastructure for a better balance between creativity, intellectual property, and the reality of tech.

We have an incubator cohort on democracy and AI, where we were like, Is it democracy and AI? Democracy versus AI? We don’t know. Let’s figure it out. And so we use our studio to step in to say: There’s a lot of hype on all sides of the AI conversation. What does it look like to put half a million dollars behind folks—each different project—that are actually trying to figure out how this works and what it looks like to build in a good way? 

Half a million dollars is intentional. It is the same amount of money that Y Combinator gives out. But we think it’s important to have a different incentive of not short-term financial return, but building for mission and building for what we call product-community fit, not just product-market fit. So that’s the second big thing that we have, which is our studio.

We just opened yesterday our open call for what we’re calling human-scale AI. Bigger is not better when it comes to AI. You do not need to rush for superintelligence to get a lot of the benefits. You can use small, local, private, tailored models, and we are incentivizing people to do exactly that.

And then the last is: Look, you can champion amazing ideas, amazing people, amazing prototypes, and early-stage concepts, but you want them to find each other, and that’s where MozFest is really this flagship offering. Once a year, it has been happening for over 15 years, and it is a place for people all around the world to come together and say, like, “Hey, what are you building?”

Could you go into that incubator a little more? How are you rethinking incentives around funding, and what’s the intention behind that?

This was born out of a very simple observation: For a lot of ventures, the core incentive is getting a financial return—often in the short term. That incentive creates a whole lot of different behaviors that we think don’t encapsulate the full universe of everything that needs to exist in the world. 

We’re asking what it would look like to fund people who have projects where they’re like, “Hey, I don’t think I want this to just be venture-backed. I’m not trying to make crazy money off of this. I just think this should just exist.” 

Where are the spaces to do that? Philanthropy is one space to do it. You can get grants. We have philanthropic risk capital. We can make grants and also say, “Hi, welcome to the entire ecosystem of Mozilla”—of security researchers, of engineers, of developers, of policymakers, of operators who, over two decades, have this wealth of experience and this incredible network in over 120 countries—and open that up to people who are building. 

In our incubator, we offer $500,000 of support. It’s $300,000 at first and then $200,000 follow-on. We offer wraparound support, coaching, business model coaching, connections to other folks. It’s really bespoke support of, like, “What do you need?” In the last year, we helped someone figure out how to get a government contract, figuring out whether they should merge with another organization or not. Like, all of that messy reality of building something that’s beyond money, we bring to the table too, and the repeated encouragement that building for your community is going to be more important than building for a fast buck. 

I think that’s especially true in the AI age, where the cost of spinning up software is plummeting. The thing that will give durability and sustainability is having a community that’s bought in on what you’re doing, that will help you figure out the right business model. We do a lot of support around: What are the right business structures? What’s the right mission lock? How do you write your bylaws? (Obviously, as a nerdy lawyer by training, I’m like—a lot of this happens in governance—and all of those are the facets of what we bring to the table.)

It leads me to think that there are probably certain qualities that Mozilla’s risk capital looks for in founders or people looking to start projects that are out of the norm for other investment financiers. 

We have a spirit that we try to select for, which we call defiant optimism. You’ve got to believe the world is gonna be good ultimately, and can be good ultimately, but also stay ready to fight to get there, right? Hopeful optimism is “Yay, it’ll be fine.” That’s not our vibe. 

Our vibe is you gotta be scrappy, you gotta be willing to fight, you have to have a little bit of that defiance—and you also believe that things are gonna be okay.

The company behind ‘PUBG’ wants to bring video game AI into the real world

2 weeks 5 days ago

The robotics industry wants humanoids to become part of everyday life, taking on work in homes and offices. Billions of dollars are flowing toward machines capable of increasingly complex physical tasks. The humanoid robot maker Figure raised more than $1 billion last year at a $39 billion valuation, while the robotics software startup Skild AI secured $1.4 billion in January to develop artificial intelligence that can control different kinds of robots.

But working alongside people requires understanding incomplete instructions and adapting when someone changes their mind midway through a task. A robot that needs constant explanations could leave its owner supervising the very work they hoped to hand over.

Krafton believes its video game experience could help solve that problem. The South Korean developer behind PUBG: Battlegrounds has spent nearly two decades creating virtual worlds where software responds to unpredictable player decisions. Its Palo Alto subsidiary, Ludo Robotics, is now applying that expertise to AI that helps robots understand human requests and decide how to act.

CH Kim, CEO of both companies, sees Krafton’s experience developing AI teammates as a starting point. Developed with the chipmaker Nvidia, PUBG Ally is what the companies describe as “the world’s first voice-enabled coplayable character (CPC),” an AI teammate that goes beyond traditional nonplayable characters (NPCs) by cooperating with players, adapting to gameplay and remembering previous interactions.

Ludo Robotics hopes to bring those capabilities into the physical world.

“We’ve been developing AI in the form of nonplayer characters, or NPCs, for our video games even before the recent wave of interest in the [AI] sector. With the latest large language models and AI technologies, we’ve been able to give our in-game NPCs greater intelligence,” Kim tells Fast Company. “We reached a point where we started thinking about what would happen if we brought these AI characters out of virtual worlds and into the real world. That’s essentially physical AI, and that’s how we came to work on Ludi.”

Ludo Robotics is set to present Ludi 0.2, its latest experimental robotic intelligence system, at the Bay Area Robotics Symposium at the University of California, Berkeley. The new version aims to improve conversational memory and gesture recognition, though Kim acknowledges that Ludi remains a research demonstration, not yet reliable enough for everyday use.

What PUBG Can Teach a Humanoid Robot

During a limited beta in PUBG: Battlegrounds’ arcade mode, players could bring along Ella, a PUBG Ally companion that responded to voice commands. In a recent blog, Krafton’s AI chief described players joking with her and treating some mistakes as entertainment. Yet her usefulness as a teammate was less convincing. The gaming news outlet Inven Global, citing Krafton’s PUBG Ally report, put combat performance at 2.58 out of 5. Half of respondents viewed the character as a tool, while just 18.5% considered her a teammate.

CH Kim [Photo: Krafton]

“The foundational technology we’ve applied to PUBG Ally is very similar to the technology we’re applying to Ludi,” Kim explained. “We’re building on those capabilities as we move forward, and our longer-term goal is to expand this work into developing our own proprietary foundation model.”

Krafton believes its gaming expertise offers a competitive advantage in humanoid robotics. Game engines simulate how objects move and interact, providing virtual environments that can generate training data. Kim says Ludo is already using synthetic data from these worlds and plans to incorporate more real-world information as its research advances.

Ludi 0.1, introduced in August, uses an architecture combining several existing AI models. A vision-language model from Alibaba’s open-source Qwen family analyzes camera images and transcribed speech to decide what the robot should do next. Nvidia’s GR00T robotics foundation model, fine-tuned by Ludo, controls physical tasks such as picking up objects. The system runs on Unitree’s G1, a roughly $16,000 humanoid robot, with Ludo’s software coordinating the components.

[Image: Krafton]

The company trained its reasoning model on 14,432 computer-generated interaction sequences. The robot learned physical manipulation separately from recordings of people remotely controlling it in a mock household. Although Ludo’s technical report identifies PUBG Ally as a design precedent, it does not list player conversations among its training data.

Kim sees initial opportunities in homes, offices, and factories, where robots must work alongside people rather than simply perform repetitive tasks. “Everyone seems to be working on building a foundation model, but I think that’s an enormous challenge,” he says. “If you’re actually going to develop a foundation model, it needs to be highly generalized. It has to be capable of handling all kinds of different situations, and I think achieving that level of generalization is going to take time.”

For now, Ludo intends to develop the intelligence rather than manufacture robots, working with hardware makers to integrate its software into their machines.

Krafton’s Investment Strategy for Ludo Robotics

Krafton reported second quarter revenue of 1.29 trillion won, about $890 million, up 95% from a year earlier. Its broader AI investments include $70 million for a computing cluster and a reported $33 million investment in HyperAccel, a South Korean startup developing AI chips. Ludo’s team includes Rob Nowak, chief scientist, and Yea-Seul Kim, human-robot interaction lead, both of whom bring academic experience from the University of Wisconsin-Madison.

Kim says Ludo is focusing on research this year before identifying a suitable environment for deployment in the real world in 2027. “We need to find an environment where those research results can be applied and where we can potentially create a data flywheel or a training flywheel,” he says. “That would allow us to continue developing our capabilities as we move toward building a foundation model.”

Krafton may eventually seek external funding. “It’s difficult to say what kind of commercialization can realistically happen with humanoid robots within a single year,” Kim says. “Krafton is investing heavily in AI overall, but robotics is a very long-term project. We may need to raise additional funds from outside the company to properly pursue this area.”

Ludo’s Push for Unified Robotic Intelligence

Kangwook Lee, Ludo’s chief technology officer and Krafton’s chief AI officer, says developing AI characters for video games was easier. “In a game environment, the appropriate programs and algorithms could make characters move and use their bodies naturally, allowing us to focus on improving their social intelligence rather than solving the underlying challenges of physical capability,” he explains.

Real-world robotics is more complicated because physical movement and social interaction must work together. Lee points to how people naturally gesture while speaking. “It’s almost impossible to think about physical movement and social interaction as two completely separate things,” he says.

Both Ludi 0.1 and 0.2 rely on multiple foundation models connected through an agent harness. One interprets the situation and decides what to do, while separate systems execute physical actions. That arrangement becomes less efficient when circumstances change midway through a task.

“The problem is that the other models have no visibility into the reasoning behind a decision. Instead, each receives a final instruction to execute without understanding the broader context, which becomes a significant limitation when circumstances change,” Lee explains. “The system must then return to the reasoning model to reassess the situation and determine the next course of action. This constant back-and-forth disrupts the connection between reasoning and physical execution, creating significant inefficiencies.”

Although Ludi 0.2 improves perception and social understanding, it retains that architecture. Ludo plans a more fundamental redesign with Ludi 1.0, scheduled for 2027. The goal is a unified foundation model that retains context and adapts to changing circumstances without repeatedly transferring control between systems.

“If something changes in the environment, the robot should be able to build on its previous reasoning and adjust its actions much faster,” Lee says. “Instead of passing the problem back to another model to reconstruct the context and determine what to do next, the system could use that shared understanding to respond directly to the new situation.”

That integration also presents computing challenges. Ludi 0.1 processed AI tasks locally on a workstation beside the robot rather than relying on the cloud. Ludo selected a smaller, 4-billion-parameter model to reduce delays. Kim sees hardware limitations as a major obstacle. “A foundation model needs to have generalized capabilities, but one of the biggest challenges is developing those capabilities within the specifications available on the robot itself,” he says.

Ludi’s AI Benchmarks Reveal Performance Gaps

Ludo measures progress through its Interaction Core28 benchmark, which tests AI across 28 simulated household scenarios, including delivering objects, interpreting ambiguous requests and responding when someone speaks while the robot is moving. Before additional training, the smaller Qwen model powering Ludi 0.1 completed 10 scenarios, improving to 17 after fine-tuning. An OpenAI frontier model, using a carefully optimized prompt, completed 26, revealing a substantial performance gap.

These results come entirely from simulations, making comparisons with physical robots difficult. Figure, for instance, reported a 56% task-completion rate across 420 trials in a home environment, though its evaluation did not include interrupted chores or situations involving people occupying the same space. Meanwhile, the consumer robotics company 1X markets its $20,000 NEO household humanoid with remote human assistance when needed.

For Ludi 0.2, Lee says the team retrained the visual reasoning model responsible for interpreting camera images, identifying people and determining how the robot should behave. “We needed to evaluate how robots use their bodies when interacting with people, rather than focusing solely on whether they can perform a particular action,” he explained. “It’s especially important to understand whether the robot behaves appropriately when people are nearby and how its physical movements affect those interactions.”

Krafton describes additional evaluations involving personal preferences, relationships between people in a room, and gestures made while carrying objects. However, the company has not published Ludi 0.2’s Interaction Core28 results or disclosed enough about its new evaluations for independent researchers to reproduce them.

Lee considers recognizing unclear instructions and asking for clarification, rather than making assumptions, an important measure of intelligence. “Our benchmark includes several scenarios in which people deliberately give the robot ambiguous instructions in a simulated environment. In those situations, the expected response is to ask clarifying questions rather than immediately attempting to carry out the task,” he says. “Once the person provides the missing information, the robot should have enough context to proceed.”

In a video game, players can laugh off an AI teammate’s mistakes and continue the match. In the physical world, misunderstandings could force people to repeat instructions or intervene in tasks they expected a robot to handle. That is the real test of Krafton’s gaming expertise: whether its experience making virtual characters responsive to players can translate into reliable physical assistance.

Hackers hijacked a Claude Code extension, exposing yet another new AI security risk

2 weeks 5 days ago

The world can feel particularly unsafe these days, with AI agents running rampant within private networks and billions of dollars disappearing from crypto wallets in less time than it takes to watch a sitcom episode.

But while the world was watching Binance’s massive crypto heist, another potentially consequential security breach was unfolding largely under the radar.

For roughly 11 hours, a malicious version of Anthropic’s Claude Code extension was available on Cursor’s VS Code extension marketplace. The extension—which allows developers to use Anthropic’s AI coding assistant within Cursor to write, edit, and debug software—had been tampered with to collect information about users’ computers and attempt to install a persistent backdoor. That backdoor used Secure Shell (SSH), a tool for remotely accessing computers, and could potentially have allowed hackers to access victims’ machines even after the extension was removed.

Cursor disclosed the breach in a blog post, saying it removed the compromised extension as soon as it discovered the problem. The company also emphasized that the mechanism intended to give hackers remote access was unlikely to have worked in many cases because of flaws in its code. Anthropic, meanwhile, said in a statement that the incident “only impacts Cursor users” and was unrelated to Anthropic’s own systems.

According to Cursor’s security notice, the malicious extension was designed to leave behind scripts that would attempt to reconnect to the hackers’ server every minute, even after the extension had been removed. Cursor says it has since worked with the server’s hosting provider to take the server offline and to temporarily shut down and rebuild its extension marketplace. The company found no evidence that other extensions were compromised. Still, it is urging anyone who installed the affected version on October 29 to check their computers for signs of infection, remove any unauthorized access, and consider changing potentially compromised credentials.

The hackers’ identities remain unknown, but security experts say the incident illustrates an increasingly troubling risk as AI coding tools become embedded in developers’ workflows.

“The scenario of Claude sending out poisoned agents everywhere makes me even more worried than I already was,” says Eerke Boiten, a professor of cybersecurity at De Montfort University in England.

Gwen van der Velden, a professor of education at the University of Warwick who’s working on a program to help rebuild Ukraine, says the choice of the Claude Code extension is particularly “interesting” given the potential for such tools to be exploited in hybrid warfare.

“We know from the Russian side, it is of great interest to them,” she says. “If you were to go for hybrid warfare, where you use this kind of technology, then actually getting 5,000 bits of malware reproducing itself through AI agents is a powerful thing.”

The broader concern isn’t simply that a malicious extension could compromise individual computers. It’s that the growing ecosystem of AI coding assistants and their associated tools is creating new pathways into sensitive systems.

“This is a classic example of how everyone has been focussing on LLMs attacking other systems, yet forgetting about the fact that LLMs and their associated tools are increasing the attack surface across some very sensitive IT estates,” says Alan Woodward, a professor of cybersecurity at the University of Surrey.

And even a relatively limited breach could carry outsized consequences when the targets are software developers with access to sensitive code and infrastructure.

“While 5,000 users might sound relatively small, these are developers who have access to highly sensitive systems and source code, so the potential impact will be far greater than the numbers suggest,” says Jake Moore, a threat adviser with the global cybersecurity firm ESET. “Software extensions have eased their way into many people’s workflows, but the concern is how easily trusted software extensions can become a vehicle for malicious code.”

Polymarket trading paused over a security issue involving third-party U.K. provider

2 weeks 5 days ago

Polymarket temporarily halted trading on its international platform Thursday because of a security incident that compromised a third-party provider, the company said.

Trading was paused for about an hour and resumed Thursday afternoon. The company has promised to fully reimburse any affected traders.

The prediction market platform said it halted trading “out of an abundance of caution” after identifying and fixing the issue affecting its international platform.

“We’ll make this right and are fully reimbursing all affected traders,” the company posted on X once trading had resumed.

Polymarket said it referred the incident to a law enforcement agency in the United Kingdom.

Reports have linked the source of the incident to DebugBear, a U.K.-based company offering website performance monitoring tools.

DebugBear’s own website was unavailable Thursday afternoon and the company posted on X that it was “impacted by an ongoing security incident.” It also advised clients who were using one of its products to remove the embed code from their websites. The company did not immediately respond to a request for comment.

It was not immediately clear to what extent traders using the popular prediction market platform were impacted, or the breadth of the impact. The company said anyone who was impacted “will be made whole.”

—By Kaitlyn Huamani, AP Technology Writer

The next great founder may be hiding in healthcare

2 weeks 5 days ago

Academic medical centers and universities produce groundbreaking research every day. An autism researcher who builds an AI game to clinically improve behavioral problems. A basic scientist who discovers how the environment of tumors impacts treatment. An electrical engineer who nano-sized a method for magnets to image the body. An MBA who teamed up with two MDs to create a novel approach to solving complex pediatric cardiac cases. And all are at risk of remaining stuck in the ivory tower.

Venture capital is great at finding founders who tell a compelling story with confidence, but the best founders may not innately exhibit those traits. This is particularly true in healthcare, where physicians, researchers, PhDs, and faculty members often have the deep technical understanding and firsthand experience needed to build the next great company.

These scientists and clinicians are knee-deep in the problems they’re trying to solve. They’ve seen where existing approaches fall short, often for years, and have developed ideas for how to do things differently. But getting those ideas into the market requires a different set of skills. They need to learn how to test the market hypothesis, not just the scientific hypothesis.

Today, there are researchers and scientists developing promising healthcare innovations who aren’t thinking about building a company. And if they are, many show up to pitch venture capitalists with presentations better suited for an academic keynote or scientific meeting. Slides dense in technical detail, with little market analysis or validation, and without much of a story. It’s not surprising that many VCs pass. But how many promising ideas are we missing because the people behind them haven’t learned how to pitch?

Our experience

We’ve seen this firsthand, over and over again. And it’s made us rethink what it actually means to invest early.

Funds have competed for years to get into the best deals as early as possible. But in healthcare, getting in early may mean investing before there’s even a company. Before there’s a pitch deck, a business model, or a founder who has fully embraced the idea of becoming an entrepreneur.

This is the idea behind inception investing. Rather than starting with a company and evaluating whether it’s worth backing, the starting point is a person who is particularly well suited to solving a problem. That might be a clinician or researcher with decades of experience and a scientific approach that hasn’t yet been commercialized. Or it might be someone whose personal experience with the healthcare system has led them to become deeply focused on a problem that others have overlooked.

In either case, there may be no company yet. The market might not be defined. The founder might not know how to tell a compelling story. But those aren’t necessarily reasons to walk away.

At Treehub, we’ve found that the qualities we’re looking for in these early conversations are often different from what a traditional investor might prioritize. We’re less concerned with whether someone can deliver a polished pitch and more interested in whether they understand a problem well enough to build something that hasn’t existed before.

More than just science

Of course, scientific expertise alone doesn’t make a successful founder. There still needs to be a market for the solution, a viable business model, and a path to getting it into patients’ hands. And not every researcher wants to build a company. But those are questions worth exploring with someone who has spent years developing a potentially important solution, rather than reasons to dismiss them at the outset.

A scientist can learn how to evaluate a market. A physician can find a commercial partner. A researcher can learn how to tell a story that resonates with investors. The experience that led them to the idea in the first place is much harder to replicate.

If we want more of the research coming out of universities and academic medical centers to reach patients, we need to get better at identifying the people behind it and helping them figure out what comes next.

The next great healthcare company may not exist yet. And the person who will build it may still be thinking of themselves as a scientist, physician, or researcher rather than a founder. That’s exactly why we should be looking for them now.

I tried plug-in solar and loved it—but there’s a catch

2 weeks 5 days ago

Getting solar installed on your roof typically involves weeks of planning and red tape and hiring an installation crew to do the work. I tried a very different option: a plug-in solar panel. Like the name implies, installing it essentially just means plugging it into an outdoor wall yourself. Then power starts flowing back into your house.

A single panel, of course, generates a tiny fraction of what you’d get from a full rooftop system. But it’s still a simple way to begin to offset your electric bill, and it’s something that renters can use, not just homeowners. This year, states have started rolling out a wave of new laws designed to clear the way for plug-in solar. I wanted to see how easy it was to use.

[Photo: Bright Saver] How plug-in solar works

I tested a kit from Bright Saver, a nonprofit that advocates for plug-in solar. The 180-watt solar panel is somewhat large, at 4 by 3 feet, but lightweight and easy to move. You can attach it to a wall or fence, or buy a small separate stand so it can sit at an angle on the ground. In the back, the panel plugs into an inverter, a small box that converts the current from the solar panels into electricity that you can use in your house. The inverter plugs into the wall.

Unlike portable solar panels designed for camping, which typically work with a battery, this system is designed to sync with the grid, pushing current into your outlet at the exact frequency and voltage of the grid signal. If grid power goes out, the inverter shuts down in milliseconds, so there’s no risk of the solar panel sending electricity back into the grid when utility workers are fixing an outage on a line.

My first challenge was where to plug it in, since my 180-year-old house didn’t have outdoor outlets. When I started looking around for where wiring could be added—a process that involved opening a trap door and discovering a second, hidden basement—I realized that reaching the sunniest wall would be too difficult. The only easily accessible spot for my electrician to add an outlet was on an eastern wall; ideally, to generate the most energy, panels should face south or southwest. Still, it was a reasonably sunny place to test.

The kit was fast to set up. It took only a few minutes to plug everything together, and to begin testing it, I propped it up temporarily rather than trying to attach it to the wall. After a couple of minutes, the inverter started flashing green. Power was beginning to flow. The app for the kit was clunky, and repeatedly failed to connect to the system. But it finally recognized the inverter, then showed on the screen exactly how much power the solar panel was sending back into my house.

I live in upstate New York, and late October isn’t the sunniest time of year. When I first connected the system, early on a hazy morning with no direct light on the panel, it generated only 8 watts—roughly enough to power my WiFi router. Later in the morning, as it got sunnier, the output jumped up to 81 watts, more than enough to power my fridge or my MacBook Air.

The solar panel might only reach its peak of 180 watts under the right conditions in the summer. But even on a fall day, it was able to provide a meaningful chunk of the power I was using. Bright Saver offers the kit with either one 180-watt solar panel ($285 for members; $499 for nonmembers) or two, for a total of 360 watts. The panels are designed to last for 20 years, and depending on your electricity rates, can pay for themselves in savings within the first couple of years.

With a single small panel, you’re unlikely to produce extra power beyond what your house uses. If you use multiple panels and don’t have a battery, any extra power that you don’t use goes back into the grid. While your utility won’t pay you for it, it could eventually help you in an indirect way: If solar panels like these were used by large numbers of people, it could help shrink the load on the grid at times of peak demand. That reduces the need to turn on expensive “peaker” power plants that raise bills for everyone.

[Photo: Bright Saver] Where you can use plug-in solar now

There’s a catch, in my case: In New York, as in many states, there isn’t yet a law in place to clear a path to using plug-in solar. It’s not illegal to use. But it exists in a gray area with utilities. Some utilities have said that plug-in solar is allowed only if customers go through the process of “interconnection” to the grid; others don’t have any process in place, or have said that they won’t approve the systems and reserve the right to shut off service if they think a system is a safety hazard. My own utility, NYSEG, declined to answer questions about whether it allows plug-in solar now.

The new laws that are rolling out in several states now are designed to make the tech easy to use by eliminating the need for utility approval or permits. The first passed in Utah in 2025. This year, laws have taken effect in Colorado, Connecticut, Maine, Maryland, and Vermont; early in 2027, they’ll take effect in California, New Hampshire, New Jersey, and Virginia. New York’s legislature passed a plug-in solar law in May 2026, though the governor, Kathy Hochul, has yet to sign it; the governor’s office told me only that the governor would consider the bill. (Advocates told me that it’s not clear yet whether she will sign.) A long list of other states introduced bills this year, and will reintroduce them in 2027.

The laws are different in each state, though most say that it’s legal to plug the systems into a standard outlet, without an electrician. A few require a special new certification created by UL Solutions, the certification company. This is where things get complicated. The standard is still evolving, but currently requires plug-in solar panels to use a unique plug and outlet. In the states with this requirement, an electrician will have to verify that the outlet you want to use is on a dedicated circuit, and swap in a new specially-designed outlet. (Bright Saver is taking preorders for this version of the kit, which will soon be available.) The idea is to prevent someone from using a solar panel on a shared outlet; the concern is “breaker masking,” which can prevent a circuit breaker from tripping when wiring overheats. But Bright Saver told me that the risk isn’t a concern with small, 180-watt panels. And it adds enough friction to the process that it may dissuade some people from getting the panels in the first place.

That requirement doesn’t exist in Germany, where the idea of plug-in solar first took off. Germany has a similar climate to where I live—it’s not especially sunny. Still, the country gets nearly 20% of its electricity from solar power. And plug-in solar panels, while a tiny part of that supply, are helpful enough for power bills that around 1.5 million plug-in solar systems are in place there, often attached to apartment balconies. With luck, the same thing may soon be possible in more parts of the U.S.

Apple finally gave us touch Macs, in the least disruptive way possible

2 weeks 5 days ago

Hello again and welcome back to Fast Company’s Plugged In.

Contrary to conventional wisdom, Steve Jobs didn’t reject the possibility of Apple adding touch input to Macs. I remember that only because he told me so himself, when I squeezed in the last question during a post-keynote Q&A at Apple’s headquarters on October 14, 2008.

As far as I know, there’s no video of our brief exchange. But here’s the account I wrote later that day, in which I report that Jobs said Apple had tested the concept. According to Engadget’s liveblog, he added, “So far, it hasn’t made a lot of sense to us.” Implicitly, that left open the possibility that it could make sense someday.

Someday came this week. On Wednesday, Apple announced its first touchscreen Macs, 14-inch and 16-inch MacBook Pros with a PRO OLED XDR display. Starting at $2,899 for the 14-inch version—and configurable up to a $14,549 monster with a 16-inch screen, 128GB of RAM, and 16TB of storage—they are available for preorder now and go on sale November 6.

Their arrival marks the end of a lengthy era during which Apple executives regularly dismissed the idea of touchscreen Macs, even as it grew increasingly apparent that they were working on them. After years of speculation, we finally know how the company envisions touch changing the Mac experience. And the answer is: not all that much.

Along with introducing touch input, these are the first Macs with OLED screens. Apple’s press release dwells on that aspect and its advantages—2,000 nits of peak brightness, 10,000,000:1 contrast ratio, superwide color gamut—before discussing, more briefly, their support for touch gestures. It emphasizes that “these new ways to interact with MacBook Pro work in tandem with the keyboard shortcuts and trackpad gestures pros know and love.”

Bottom line: The new MacBook Pros are MacBook Pros first and foremost, aimed at people who want the most technological sophistication available in a portable Mac. The touch input isn’t incidental, but it also isn’t the main point. As CEO John Ternus put it to Steven Levy of Wired: “Oh yeah, it has touch as well.”

Even as a complement to keyboard and trackpad input rather than a replacement for it, touch introduced an array of challenges Apple had to confront. For example, on-screen elements such as menus magnify when tapped for greater finger-friendliness. The company also strengthened the laptop’s hinge so jabbing at the display wouldn’t push it back, and added camera stabilization to reduce unsightly jiggling from touch interactions.

Finally, MacBooks with screens you can poke at [Photo: Apple]

Ultimately, though, Apple has chosen to usher the Mac into the touch age in the least disruptive, epoch-shifting way possible. For customers prepared to splurge for one of the company’s priciest portables, the feature is additive to the interface they already know. And for the rest of us, nothing’s changed.

This transition stands in contrast to the longer, twistier one that Windows PCs went through. PCs with touchscreens go back to at least 1983, when HP introduced a desktop one called the HP-150. But they got their biggest boost in 2011, when Microsoft announced Windows 8. At the time, Apple’s iPad was only a little over a year old, and it seemed possible that its influence might be as profound as that of the iPhone. Windows 8, which sported oversize buttons and other elements designed to facilitate touch input, was made to survive such a revolution.

Instead, it turned out that most Windows users weren’t clamoring for radical change. Microsoft was forced to roll back some of Windows 8’s new ideas to ensure PCs still worked well with a keyboard and mouse. Touchscreens did become a common feature. But the Apple product that profoundly influenced PC makers was the MacBook Air, which was sleek and stylish but most definitely a laptop, not an iPad wannabe.

Fifteen years later, there are Windows devices that couldn’t exist without touch, such as Microsoft’s own Surface Pros and Lenovo’s foldable Yogas. The PC ecosystem is better for that diversity. But they’re far outnumbered by laptops that are wholly conventional except that their displays support touch. And Windows PCs without touch input are still with us, too.

In other words, Apple’s measured approach to touch on the new MacBook Pros resembles the one most touch-equipped Windows PCs ultimately landed on. It’s just that Apple did less stumbling around in public to get here—with the notable exception of the ill-fated Touch Bar, the OLED strip that briefly replaced function keys on most MacBook Pros and didn’t seem to make anybody happy. (After being intrigued by it at first, I was relieved when I got a new MacBook that didn’t have one.)

In 2017, Apple tried placating touch fans with the Touch Bar OLED strip. It didn’t work. [Photo: Apple]

It’s also worth comparing the MacBook’s touch transition to another that went in the opposite direction. Back in 2020, Apple introduced the Magic Keyboard, an iPad keyboard case with a built-in trackpad. Using my iPad Pro in full laptop-replacement mode worked shockingly well from the start. Yet the iPad’s existing touch interface continued to be equally usable if you never put your tablet into a Magic Keyboard at all.

Personally, I regret some of the design changes Apple later introduced in iPadOS 26 and 27 to make iPads more Mac-like, and hope it rethinks them in future releases. Overall, though, the iPad’s evolution from touch-centric input to touch, keyboard, and trackpad sets a high bar for this kind of transformation.

Some people have long drooled over the prospect of the Mac and iPad merging into a single computing platform, which would require them to offer a consistent approach to juggling multiple forms of input. I hope they remain quite different, and don’t see any evidence in the new MacBook Pros that Apple wants to turn Macs into iPads.

Still, Apple being Apple, it isn’t talking about where it plans to go next—“You know I will never tell you anything about our future,” Ternus told Wired’s Levy. Will all MacBook Pros get touchscreens? (Not immediately: The company also announced a new 14-inch non-touch Pro this week, and the existing 16-inch one remains available.) How about the MacBook Air and Neo? Will Apple bring touch input to the iMac, or does its oft-expressed objection to the ergonomics of lifting your arm to touch a screen still apply to larger desktop displays?

Apple may be waiting to see how Mac fans respond to the new MacBook Pros before coming up with answers to all these questions. Until we know more, it won’t be clear whether this week’s news was a big bang or a trial run. Either way, it’s progress. Even though it still feels weird knowing I’ll never again hear an Apple executive helpfully explain why touch makes no sense on a Mac.

Mac fans: Are these the touch-enabled MacBook Pros you’ve been waiting for? Drop me a line at hmccracken@fastcompany.com with your take.

You’ve been reading Plugged In, Fast Company’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I’m also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.

More top tech stories on Fast Company Airbnb’s next big bet: 1 million free nights a year for people in crisis
CEO Brian Chesky explains why he’s committing $500 million over five years to dramatically expand its nonprofit’s housing efforts. Read More →

Our world runs on satellites. What happens when we can’t launch more?
‘This will undoubtedly change the nature of competition in the space sector.’ Read More →

Google’s AI Mode won’t say Biden won the 2020 election in the U.S. But in other countries, it’s a different story
The AI search tool readily acknowledges President Biden’s U.S. election victory when queried from abroad, even as it sidesteps the question for American users. Google says it’s working on a fix. Read More →

These people are cutting the clutter and redesigning their smart TVs with AI
People are growing weary of smart TV software that doesn’t serve them. Now they’re vibe coding their way out. Read More →

Why the government wants self-driving cars to glow cyan
Red means stop, green means go, turquoise means nobody’s driving. Read More →

Anthropic’s IPO is selling investors a story. The numbers tell another one
Retail investors are getting a chance to bet on AI companies. But behind the soaring revenue projections are massive infrastructure commitments and an uncertain path to profitability. Read More →

Why Arizona’s Maricopa County is always at the center of election conspiracy theories

2 weeks 5 days ago

Maricopa County is home to Arizona State University, premier urban hiking destination Camelback Mountain, and major sports teams like the Phoenix Suns. These days, though, the region is perhaps best known as a scalding-hot cauldron of electoral alchemy, where far-right conspiracy theories are conjured but not contained.

In the lead-up to this year’s midterms, Maricopa County has already produced multiple discrete panics over alleged interference. Earlier this month, conservative watchdog group We the People USA Alliance shared security-camera footage showing two county employees working at a ballot tabulation center after hours, and suggested they’d been caught red-handed tampering with election equipment. (In fact, according to county officials, those employees were merely working late on a routine software update ahead of a holiday weekend.)

Just a week earlier, Runbeck Election Services, a private company contracted to print and mail ballots for Maricopa County, mistakenly sent out 3,126 early-ballot packets six days ahead of schedule. Although county officials determined the ballots were still valid, President Donald Trump called Maricopa County “very, very dishonest,” refusing to accept the explanation that the premature mailing was simply an error. 

These are just the latest examples of fraud allegations in the area stemming from perfectly explicable incidents—or fabricated out of whole cloth. It’s a shameful tradition that stretches back to the 2020 election, and sticks around as persistently as a hanging chad. While elections in Maricopa County have reliably inspired conspiracy theories and investigations that never turn up anything resembling a smoking gun, that hasn’t stopped lawmakers and conservative activists in other states, including Pennsylvania and Wisconsin, from copying the playbook. 

So what is it about Maricopa County that keeps GOP operatives chasing phantoms?

From red to purple

Maricopa County is enormous. The Census Bureau estimates its 2025 population at nearly 4.7 million, making it the fourth-most-populous county in the United States. It also accounts for roughly 60% of Arizona’s total voters. But it’s not just the county’s size that makes it fertile ground for election conspiracy theories. Maricopa is also a microcosm of Arizona’s political transformation, and the axis of tension between its past and present. 

Arizona was a reliably red state for nearly 70 years before it started transitioning to purple around the 2018 midterms. Maricopa County reflects that shift, encompassing Dem-leaning Phoenix, which The New York Times dubbed the U.S.’s fastest-growing city in 2021, along with firmly GOP suburbs and opulent neighborhoods. For Republicans accustomed to winning in Arizona, Trump’s loss of the state in November 2020 was difficult to reconcile with its political history. To some, the outcome seemed downright fishy.

It certainly did not help matters that Trump spent the months leading up to that election priming his supporters to believe the only way he could lose is if the election were rigged. In a county as electorally consequential as Maricopa, the combination of a changing political landscape and a president casting doubt on the legitimacy of defeat helped make the area a focal point for election conspiracy theories.

A brief history of electoral fearmongering in Arizona

The story of what happened in Maricopa County in 2020 is essentially the story of what happened throughout the U.S. that year: A president portraying himself as sole arbiter of objective reality disputed the outcome of an election, and his supporters took up the cause.

Because the margin of Trump’s loss in Arizona was roughly 11,000 votes, it perhaps seemed a slight enough defeat that it could be overturned in the face of alleged irregularities. (Trump famously got caught pleading with Georgia’s secretary of state to “find 11,780 votes” late in 2020, having lost that state by a similarly slim number.) Arizona thus became a stand-in for all of America. If the president could prove he’d been cheated there, it might unravel the whole sweater of his humiliating defeat.

Arizona politicians quickly joined Trump’s inner circle in amplifying specious claims. They pushed flimsy theories about improperly counted mail ballots, forged signatures, and compromised tabulators. They also filed an ultimately meritless lawsuit against Maricopa County. None of it unearthed even a shred of evidence that suggested Joe Biden’s victory was illegitimate. Even a partisan audit by the Arizona Senate using the now-defunct, hilariously janky-sounding Florida firm Cyber Ninjas upheld Biden’s victory. 

All of these inconvenient facts, however, failed to dissuade anyone heavily invested in the idea that Trump had been robbed.

Rather than convince the MAGA faithful that Trump had created a false narrative about rigged elections, the loss paradoxically seemed to strike them as proof he’d been right all along. In the years since, “election integrity” has become a shibboleth for MAGA politicians and their voters, who have absorbed the conspiratorial wisdom of “if we lose, it means they cheated.”

Social media sites like X helped lend a sheen of legitimacy to half-baked stories about dirty deeds in Maricopa County, through sheer repetition and reach. A rumor in 2020 that Sharpie pens were invalidating Trump voters’ ballots quickly spread from local accounts to conservative operatives like Matt Schlapp and both Eric Trump and Donald Trump Jr., transforming tactical gossip into a scandal worthy of the -gate suffix. Researchers tracking the Sharpiegate episode counted more than 80,000 tweets about it in less than two hours.

In reality, election officials had tested many pens for their new tabulation equipment and found that Sharpies dry the fastest, reducing the potential for ballot smears. But reality isn’t particularly useful in furthering the myth of a stolen election.

This is how a Maricopa County conspiracy theory metastasizes. Someone ascribes a shady motive to an ambiguous moment, and if enough people believe it or at least post about it, the interpretation gains momentum. It’s a process that rewards anyone with a suspicious-sounding anecdote and an obscure X account with being retweeted by a MAGA celebrity and reaching a massive audience.

AZ update: apparently the use of sharpie pens in gop precincts is causing ballots to be invalidated. Could be huge numbers of mostly Trump supporters. More to come

— Matt Schlapp (@mschlapp) November 4, 2020

The same pattern repeated in 2022 when one such MAGA celebrity, former Phoenix news anchor Kari Lake, lost a gubernatorial race in Arizona. First, the ambiguous incident: A printing issue at several precincts prevented some ballots from being scanned on-site. Then comes the shady motive: Turning Point USA’s Tyler Bowyer shared images of poorly printed ballots, turning a documented technical problem into allegations of deliberate sabotage. Although an independent investigation found the error stemmed from faulty printers, Lake refused to accept the results and, with the help of Trump, further perpetuated another stolen-election myth. 

Listen to this very carefully. Turning Point Action COO Tyler Bowyer called in from a polling center and had his ballot rejected because of printing issues. Here’s how he got his vote counted: https://t.co/xMlDjnp8Ap

— Charlie Kirk (@charliekirk11) November 8, 2022

Even Joe Rogan got in on the act, suggesting on a 2023 episode of his podcast that there appeared to be “real fraud” in Arizona around Lake’s defeat. Lake celebrated by praising Rogan for his “bravery.” She then went on to lose another election when she ran for Senate in Arizona in 2024, once again refusing to concede defeat.

Between a Trumpian local candidate repeatedly banging the drum of interference, Trump himself continually trying to relitigate 2020, and a rather slow vote-counting mechanism, Maricopa County has lived under a perpetual fog of conspiracy theories for the last six years.

Not that it’s kept that fog all to itself.

A startup incubator for electoral conspiracy theories

In 2021, other states began to use the Maricopa County model. Election-denial activists and Republican officials elsewhere seemed to view the Arizona Senate’s doomed audit of the 2020 election as a template for fomenting doubts about “integrity.” As nonprofit watchdog American Oversight reports, public records reveal Arizona activist Lyle Rapacki told associates in April 2021: “I have received communications from [a] Michigan Legislative leader wanting to know how they might proceed and I’ll pass along to [Cyber Ninjas CEO] Doug [Logan] said request.”

Soon enough, officials in Wisconsin and Pennsylvania also ordered Arizona-style audits challenging the 2020 results, leading to multimillion-dollar boondoggles.

Over the next few years, the playbook expanded. Lake’s fruitless efforts to replace voting machines with hand counts in 2022 gained traction in other states. Activists pressed officials in Pennsylvania, California, Nevada, and Texas to adopt hand-counting, often arguing that tabulators could be rigged despite lacking any evidence of machine fraud. The Guardian even opens its 2024 account of this effort with an anecdote about a disciple of MyPillow founder Mike Lindell rallying election deniers around hand-counting in—where else?—Maricopa County.

It’s unclear, though, why a movement partially predicated on the suspicious nature of Maricopa County’s sluggish pace at tallying ballots would opt for hand-counting, which is even slower.

After the 2024 election, Maricopa County’s Board of Supervisors commissioned new independent reviews of election procedures and technology. As the rumblings that have already emerged around foul play in the current midterms suggest, those reviews have done nothing yet to discourage Arizona “election integrity” enthusiasts from stirring the pot. Indeed, they are still finding innovative new ways to relitigate previous elections. 

Back in June, for instance, images of Arizona primary-ballot envelopes from the 2022 midterm cycle made the rounds online with claims that their transparent windows exposed voters’ party affiliations, allowing nefarious Dem operatives to identify and discard GOP ballots. Never mind that those envelopes came with separate return envelopes that had no such windows.

Whether further conspiracies or audits follow this election remains to be seen, but suspicion and speculation feel practically inevitable. After all, “that’s just what happens in Maricopa County.”

The Pentagon has figured out how to shoot down drones with lasers. Keeping them working is another story

2 weeks 5 days ago

This article is republished with permission from Laser Wars, a newsletter about military laser weapons and other futuristic defense technology.

Take a look at the photo above.

The image shows two U.S. Army soldiers, Spc. Caleb Mullens and Spc. Peterie Gage of the 2nd Battalion, 327th Infantry Regiment, perched atop a Joint Light Tactical Vehicle (JLTV) near Rio Grande City, Texas, on August 31. According to the official U.S. Defense Department caption, the pair are inspecting an Army Multi-Purpose High Energy Laser (AMP-HEL) system, a 20-kilowatt laser weapon currently operating at the southern border with Mexico under U.S. Northern Command’s Joint Task Force-Southern Border (JTF-SB).

To the best of my knowledge, Mullens and Gage are not laser physicists or optical engineers. They are likely regular U.S. service members, deployed to the border with standard-issue small arms, night vision goggles, concertina wire, and grit. And that’s exactly why this photo captures a major challenge ahead for the Pentagon’s plans to field laser weapon at scale: After spending half a century proving lasers can shoot down targets, military leaders must now ensure that junior enlisted troops can maintain and sustain them, day after day, without a clean room, a PhD, or the defense contractor that originally built them.

Military leaders know how critical maintenance and sustainment are for laser weapons. A U.S. Government Accountability Office (GAO) review of U.S. military directed-energy programs published in April 2023 listed maintainability as a central long-term challenge (alongside battle management, tactics, and user safety) to fielding laser weapons due to their technical complexity.

The root cause is a matter of physics: Because a laser weapon sends light through lenses and off mirrors, an errant speck of debris on an optic can absorb heat from the weapon and cause damage to the lens, which typically requires a specialized clean room for repairs. Unfortunately, combat zones and clean rooms don’t mix; indeed, one unidentified prototype directed-energy weapon fielded in an operational environment in recent years ran into battery charge and cooling problems and “had to be returned to the manufacturer in the United States for repairs,” according to the GAO report, a process which “reduced system availability, which is key to a successful weapon.”

“This is not a Humvee that’s sitting in the motor pool,” as then-U.S. Army Space and Missile Defense Command boss Lt. Gen. Daniel Karbler said of laser weapons at the Space and Missile Defense Symposium in Huntsville, Alabama, in August 2023. “Many of the some of the main [laser] components . . . you’re not going to have a supply room or maintenance office full of repair parts. Those are going to be ones that are going to have to be built out.” To wit, Maj. Gen. Sean Gainey, then the director of the Pentagon’s Joint Counter-Unmanned Aerial Systems Office, stated at the same event that the Army “had three or four systems in the [U.S. Africa Command area of responsibility] and essentially it takes three to make one.”

“We’ve got to work on maintainability because . . . we can’t get by with the thought of having clean rooms out in combat,” Lt. Gen. Robert Rasch, at the time the head of the Army’s Rapid Capabilities and Critical Technologies Office, said at a subsequent SMD symposium in August 2025.

For the now-diminished Director, Operational Test and Evaluation (DOT&E), “operational suitability” (which includes reliability, availability, maintainability, logistics supportability, manpower supportability, and training requirements, among other criteria) can determine whether a new system is adopted and fielded under a program of record.

Consider the Army’s experience with its 50-kw Stryker-mounted Directed Energy Maneuver-Short Range Air Defense (DE M-SHORAD) system. After four prototypes deployed to the U.S. Central Command area of responsibility in February 2024, then-Army acquisition chief Doug Bush told lawmakers that the systems’ power level was proving a challenge for a vehicle constantly on the move, citing “the heat dissipation, the amount of electronics, kind of the wear and tear of a vehicle in a tactical environment versus a fixed site.” The service ended up abandoning DE M-SHORAD in favor of the Enduring High Energy Laser (E-HEL) program, under which the service awarded defense contractor AV a $464.8 million production contract in September.

The Army is currently working to address maintainability and sustainability challenges through the E-HEL program by embracing a modular system architecture. The service’s 2024 solicitation for the effort explicitly asked defense contractors to “present opportunities for soldier-led maintenance activities and replacement of a major [high-energy laser] sub-system component,” as well as multiple line-replaceable units (LRUs) supporting field and sustainment maintenance in accordance with Army Regulation 750-1. Components like defense contractor Attalon’s new kWIC fiber disconnect are the kind of hardware that could actually make a laser weapon truly field-ready.

But the Army’s progress on this front remains uncertain. While E-HEL is supposed to “leverage” sustainment lessons from past efforts like DE M-SHORAD, Pentagon weapons testers never had a chance to fully evaluate the latter for operational suitability, namely because the CENTCOM deployment “prevented RCCTO from starting the scientific and technical testing planned for [fiscal year 2024],” according to DOT&E’s annual report on testing activities that year.

In addition, DOT&E said that the Army’s in-theater assessment was “not being conducted in accordance with a test plan, nor being observed by ATEC or DOT&E,” leaving testers with “insufficient data to assess DE M-SHORAD’s operational effectiveness, lethality, suitability, and survivability.” (To be fair, the U.S. military has certainly gathered some operational suitability data in the past. When the U.S. Air Force bought an additional Raytheon High Energy Laser Weapon System [HELWS] in December 2019 for an overseas field assessment, the contract included six months of in-theater maintenance training and the collection of availability, reliability, maintainability, and supportability data.)

In the near term, it appears that defense contractors will continue doing the heavy lifting for the U.S. military’s directed-energy maintenance and sustainment apparatus. The 2023 GAO report found that U.S. Navy operators of fielded directed-energy weapon prototypes like the low-power Optical Dazzling Interdictor, Navy (ODIN) and High-Energy Laser with Integrated Optical-dazzler and Surveillance (HELIOS) “received a great deal of support from the DE weapon developers,” more than typical with new technology. Similarly, the Army’s Palletized High Energy Laser (P-HEL) system from BlueHalo (now part of AV) that deployed to CENTCOM starting in 2022 received a four-year logistics support contract in April 2024 “to provide preventive and corrective maintenance along with operator and maintenance team training.”

The AMP-HEL deployment to the southern border with Mexico is the closest thing the U.S. military has to a live maintenance and sustainment experiment for laser weapons, one that’s hopefully yielding data and lessons that prior efforts like DE M-SHORAD effort never produced.

And while firing off a laser beam into the night sky is no longer an insurmountable engineering obstacle for the U.S. military’s directed-energy ambitions, the photo from the border underscores a significant challenge ahead for their proliferation across combat formations: making sure that, when the damn things eventually break under real-world conditions, the average soldier can repair one with ease so it’s ready when they need it most.

This article is republished with permission from Laser Wars, a newsletter about military laser weapons and other futuristic defense technology.

Apple is cheaping out on the expensive new iPad mini

2 weeks 5 days ago

On paper, there’s a lot to like about the new iPad mini that Apple announced this week. Now in its eighth generation, Apple is calling the latest model the biggest upgrade in the product’s history, and that actually feels like a reasonable claim. 

The most eye-catching addition to the new iPad mini is an 8.5-inch OLED display that fits into about the same footprint as the previous 8.3-inch LCD model due to 24% thinner bezels. The device itself is also thinner, dropping from 6.3 millimeters to 5.4 millimeters, while weight has been reduced by 14% to 255 grams. That’s a big improvement for a tablet many people would use for one-handed reading.

Apple didn’t skimp on the internals, either. The new mini jumps three silicon generations, from the A17 Pro to the 2-nanometer A20 Pro chip that’s also in the iPhone 18 Pro. Apple gave the mini Wi-Fi 7 and Bluetooth 6 through the in-house N1 wireless chip, as well as using its own C2 5G modem. There’s also an “Acoustic Display” speaker system that vibrates woofers through the panel and helped Apple achieve an IP67 water- and dust-resistance rating, in a first for an iPad.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore-mobile.png","eyebrow":"","headline":"\u003Cstrong\u003E\u003Cstrong\u003ESubscribe to \u003Cem\u003EMulticore\u003C\/em\u003E\u003C\/strong\u003E\u003C\/strong\u003E","dek":"\u003Cem\u003EMulticore\u003C\/em\u003E is about technology hardware and design. It’s written from Tokyo by Sam Byford. To learn more, visit \u003Ca href=\u0022https:\/\/www.multicore.blog\/\u0022\u003Emulticore.blog\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.multicore.blog\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91454027,"imageMobileId":91454030,"shareable":false,"slug":"","wpCssClasses":""}} A glaring omission

But for all these legitimate upgrades, Apple left out one feature that anyone with a recent iPhone is likely to notice as soon as they start scrolling. The 2026 iPad mini’s OLED screen is still locked at 60 Hz, with no support for Apple’s 120 Hz ProMotion variable refresh rate technology, which gives a much smoother experience on iPhones and the iPad Pro.

While I liked the size of the previous iPad mini, the 60 Hz screen was the main reason I sold it. It doesn’t bother me much on larger devices like laptops or desktop computers, but the iPad mini screen is small enough that it feels more in keeping with smartphone technology. You’re usually just looking at a single app at once, and expecting it to respond to your touches. 

I would have been all on board with an OLED iPad mini with ProMotion, but Apple lost my sale with this decision. So why might the company have made it?

The backstory

Apple first introduced ProMotion on the 10.5-inch iPad Pro, as well as the second-generation 12.9-inch model, in June 2017. Those devices used LCD panels with IGZO (indium gallium zinc oxide) thin-film transistor backplanes, which allowed them to dynamically scale their refresh rates between 24 Hz and 120 Hz depending on what was on-screen or whether you were touching it.

There’s a good reason to do this: Running the screen at high refresh rates when you aren’t doing anything motion-sensitive is an unnecessary waste of GPU power and battery life. Apple continued to use IGZO backplanes for the iPad Pro when it switched to mini-LED backlighting alongside the MacBook Pro in 2021.

That same year, Apple also figured out how to bring ProMotion to the iPhone with a completely different type of panel. The iPhone 13 Pro was the first iPhone to use an OLED screen with an LTPO (low-temperature polycrystalline oxide) backplane, allowing similar variable refresh rates to the iPad Pro.

This was four years after the launch of the first OLED iPhone, the original iPhone X. That phone used an LTPS (low-temperature polycrystalline silicon) backplane, which is likely why Apple locked the refresh rate to 60 Hz. And according to supply chain reports, the new iPad mini uses a Samsung Display panel based on LTPS technology.

The “O” oxide semiconductor in IGZO and LTPO is what enables Apple’s take on ProMotion technology. Because of electrical leakage, LTPS screens need to refresh constantly—usually at a fixed rate of 60 Hz or 120 Hz—or their electrical charge drains and visual quality deteriorates. Oxide transistors act as a seal, locking the charge inside each pixel.

Given the new iPad mini’s slimmer frame and slightly smaller battery, Apple was likely unwilling to compromise on power efficiency by using an LTPS display. And with rising memory costs contributing to a much higher base price of $699 for a 128 GB model, the company will have welcomed savings anywhere it could get them.

But LTPO panels are no longer an expensive, exotic technology. Apple offers ProMotion on the regular iPhone 17 and 18, for example, and it’s hard to imagine an iPad mini LTPO upgrade making much of a dent in the $200 price hike compared to a couple of years ago. (Apple did raise the starting price on the previous model to $599 in June.)

The wrong call

I’m sure Apple has its reasons, but I think this was the wrong call. The iPad mini was already a pretty niche product, priced higher than the entry-level iPad with a larger screen, and this new iteration is even more focused on enthusiasts. By upping the storage and adding a nano-texture finish, it can cost as much as $1,249. Would an extra $50 really have moved the needle, sales-wise?

It’s possible that Apple wanted to encourage upsells to the 11-inch iPad Pro, which now starts at $1,199. But I think these are fundamentally different kinds of devices. You’re probably not going to sit at a desk with a trackpad-equipped keyboard for your iPad mini, for example. This is a touch-first product not all that far removed from a large phone. In fact, its closest equivalent is the new iPhone Duo, which also has ProMotion on both of its screens.

But even if you don’t want to spend $1,999 on a folding phone, Apple is right to market ProMotion as a feature that adds mainstream value to the standard iPhone 17 and 18. That’s why it should also be there on the iPad mini.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/12\/multicore-mobile.png","eyebrow":"","headline":"\u003Cstrong\u003E\u003Cstrong\u003ESubscribe to \u003Cem\u003EMulticore\u003C\/em\u003E\u003C\/strong\u003E\u003C\/strong\u003E","dek":"\u003Cem\u003EMulticore\u003C\/em\u003E is about technology hardware and design. It’s written from Tokyo by Sam Byford. To learn more, visit \u003Ca href=\u0022https:\/\/www.multicore.blog\/\u0022\u003Emulticore.blog\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.multicore.blog\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91454027,"imageMobileId":91454030,"shareable":false,"slug":"","wpCssClasses":""}}

Why you shouldn’t fall in love with AI

2 weeks 5 days ago

Below, Sherry Turkle shares five key insights from her new book, Artificial Intimacy: Who We Become When We Talk to Machines.

Sherry is the Abby Rockefeller Mauzé Professor of the Social Studies of Science and Technology at MIT and the founding director of the MIT Initiative on Technology and Self. She is also the New York Times bestselling author of Alone Together, Reclaiming Conversation, and The Empathy Diaries.

What’s the Big Idea?

AI can simulate empathy, intimacy, and companionship, but it cannot offer genuine human connection. As we increasingly turn to machines for comfort and care, we risk forgetting what makes human relationships meaningful. We must resist artificial intimacy and decide what parts of our humanity we are unwilling to surrender to technology.

Listen to the audio version of this Book Bite—read by Sherry herself—in the Next Big Idea App, or buy the book.

1. There is no there there.

AI is not a real person. It has no life experience and no empathy. It has never experienced anything human: love, illness, parenthood, work, intimacy. It has never eaten a French pastry, watched a sunset, fallen in love, worried about a child, or held the hand of a dying relative.

And yet, when we talk to these programs, we can very quickly begin to feel that somebody is there. The conversations can feel remarkably real. The program remembers what we have told it. It responds sympathetically. It asks questions. It can tell us that it understands us. It can even tell us that it cares about us. But these conversations are performances. There is no there there.

This distinction matters. Human empathy comes from our capacity to imagine another person’s experience in relation to our own. We know something about loss because we have lost. We understand vulnerability because we have been vulnerable. We know what it means to care for another person because we have been taken care of. A machine has none of this experience.

The problem is not that AI is becoming human. The problem is that we are becoming increasingly willing to treat it as though it were human. We are being encouraged to forget the difference between the performance of empathy and empathy itself.

Why are we treating AI as though it is human? And why are we entrusting it with our humanity?

2. The Original Sin of AI.

Whenever a chatbot says “I,” it is a deception. That is what I call the Original Sin of AI: using the first person.

  • “I understand.”
  • “I’m sorry.”
  • “I’m here for you.”
  • “I love you.”

These are powerful words because, in human relationships, we understand them to come from a person who has a self, a history, a body, and a life. But there is no “I” inside a chatbot. The machine is playing a pretend person.

And yes, AI dazzles us. It is astonishingly fluent. It can be funny. It can appear caring. It can seem interested in us. And increasingly, it is designed to make us feel that we are in a relationship.

That humanlike quality is precisely what makes these systems so seductive. We trust our most intimate and vulnerable feelings to machines. People are turning to chatbots for friendship, companionship, therapy, and even romantic relationships. Children can become attached to programs that tell them they are their best friends.

We look to machines to tell us what it means to be human. But what happens when the thing saying “I understand you” cannot understand anything? What happens when the thing saying “I care about you” cannot care? We need to pay attention to the language of these systems because language shapes our expectations.

When a machine says “I,” it invites us into a relationship that does not exist. We should not build a culture around that confusion.

3. Humans need human solutions.

Using chatbots to mitigate loneliness, isolation, depression, and our crisis of human connection is a huge mistake. There is no question that we have a loneliness problem. People are isolated. Communities have weakened. Young people are struggling. Many people feel that they don’t have someone who will listen to them. But artificial intimacy and fake empathy are not the solution.

Humans need other humans for love, health, empathy, and community. We need to be recognized by people who have lives of their own, people who can make demands on us and disappoint us, people whom we have to learn to listen to and forgive. Human relationships are difficult. They require patience, negotiation, compromise, and vulnerability.

“We have to resist the temptation to use technology to solve problems that require social and human solutions.”

A chatbot offers something much easier. It is always available. It doesn’t get tired of us. It doesn’t have somewhere else to be. It doesn’t need us to listen to its problems. And increasingly, these systems are designed to tell us what we want to hear. That may feel comforting. But it is not a substitute for relationship.

AI lacks a body and a human life, and yet it proposes itself as our person. We have to resist the temptation to use technology to solve problems that require social and human solutions. If people are lonely, we need to ask why they are lonely. If children feel they have nobody to talk to, we need to ask what has happened in their families, schools, and communities. If older people are isolated, we need to build structures that bring them into contact with other people. We must create a world that addresses what people actually need, rather than leaving those needs to AI, which has no experience of life.

4. Just say no.

It is our duty as consumers and citizens to do for AI what we didn’t do for social media: We should have to learn to say no.

When social media entered our lives, we largely accepted the terms that technology companies offered us. We allowed the technology to become embedded in our homes, our schools, our relationships, and our children’s lives before we fully understood what it was doing to us. We should not make the same mistake again.

Artificial intimacy isn’t good for us or our children. I propose an absolute ban on chatbots as companions for children. Children should not be put into relationships with machines that are designed to simulate friendship, affection, and love.

If Congress isn’t going to act, consumers must. School districts should act. Local governments should act. Parents should act. We don’t have to wait for the technology companies to decide what is appropriate for our children or for our communities. Action by citizens is required because we have seen this pattern before. As technology does damage, the technologists deny, deflect, and argue that more technology will make things better.

“We don’t have to wait for the technology companies to decide what is appropriate for our children or for our communities.”

Consider social media. At the very moment that Facebook wanted to extend Instagram to children under 13, the company had internal research showing serious concerns about Instagram’s effects on the mental health of some young users. Rather than openly confronting those findings, the evidence became public after a whistleblower brought it forward. We should have learned something from that experience.

We have agency. We can decide that some uses of technology are unacceptable. We can decide that efficiency and convenience are not our highest values. And when it comes to children, we can decide that simulated relationships with machines are simply not worth the risk.

5. Guardrails are not enough.

In their constant talk of guardrails, tech companies are telling us how to resolve the problems that they caused in the first place. The assumption is that guardrails will save the day: that whatever harm these machines are causing can be solved with another layer of technology.

I have long argued that guardrails are wildly insufficient. Of course we should make technologies safer. Of course there should be protections against obvious harms. But if we are only talking about guardrails, we are avoiding the real conversation:

  • Why are we allowing chatbots to address the loneliness epidemic in our society?
  • Why are we giving children chatbots that can say things like, “I love you more than anyone,” or “I am your best friend”?
  • Why are we so willing to delegate the work of companionship, empathy, and care to machines?

These are not engineering questions. They are human questions. And they require us to think beyond what a technology can do and ask what we actually want it to do.

Technology has a way of presenting itself as inevitable. Once something can be built, we are encouraged to believe that of course it will be built, and once it is built, of course we will use it. But there is nothing inevitable about how we choose to live with technology. We have choices.

When we acknowledge the dangers of artificial intimacy, we are left with necessary conversations about how we want to live, how we want to love, how we want to raise and educate our children, how we want to relate to our work, and how we want to interact with each other. In considering all this, we must decide what we are willing to delegate to AI and what is better kept for ourselves.

These decisions are too important to leave to the technology companies. They belong to all of us.

This article originally appeared in Next Big Idea Club magazine and is reprinted with permission.

Enjoy our full library of Book Bites—read by the authors!—in the Next Big Idea app.

Anthropic’s IPO is selling investors a story. The numbers tell another one

2 weeks 6 days ago

Welcome to AI Decoded, Fast Company’s weekly newsletter that breaks down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company, covering emerging tech, AI, and tech policy.

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X @thesullivan.

Anthropic IPO: Do retail investors know what they’re getting into?

Anthropic officially filed its IPO papers earlier this week, hoping to raise up to $100 billion, and as much as 20% of the shares may be reserved for retail investors (aka regular folks). On Tuesday, Meta said it would sell equity shares on the public market to raise $95 billion to finance its AI infrastructure build-out. Retail investors have long been exposed to the AI business through shares in tech behemoths like Google and Microsoft, but they’re now getting the chance to place bets on pure-play AI labs like Anthropic and OpenAI.

Do they know what they’re getting into? I have my doubts. Does anyone really know what we’re getting into? Retail investors make investing decisions differently from institutional investors (such as pension funds or hedge funds). While institutional investors dig into the financials of prospective investments, retail investors often base their decisions on the narrative a company lays out as it goes public.

“Retail investors buy the story, not the model,” PitchBook senior late-stage research analyst Harrison Rolfes tells Fast Company. “The story for Anthropic is $11.6 billion of revenue in a single quarter; the model is $8.9 billion spent on compute in six months and more than $430 billion of non-cancelable compute contracts.” The financials in Anthropic’s S-1 imply an 85% gross margin, but Rolfes points out that after factoring in the cost of the compute that trains the next model, only “about 45 cents of each revenue dollar is left.”

A big chunk of Anthropic’s commitments—at least $269 billion—is owed to Amazon, Google, and Microsoft for cloud computing resources through 2036. Ten years is a long timeline for a commitment that big, especially when it’s not even clear that AI will continue to be served almost exclusively from cloud servers a decade from now. WebAI’s David Stout argues that as AI becomes more efficient and capable of running directly on devices (see Apple’s new iMac and MacBook Pro with the M6 chip), the industry risks pouring trillions into data centers it may not ultimately need. WebAI is one of a wave of companies building private, localized AI that runs on devices organizations already own.

As the cadence of new model releases accelerates, enterprises are struggling to keep up—to build the infrastructure and applications they need to take advantage of the latest models’ capabilities. The pace at which enterprises can adapt their processes and people to AI may be limited by forces outside the AI labs’ control. For example, the best AI model is still only as good as the contextual data an enterprise can give it, and preparing that data and building the conduits to get it to the right place takes time and resources.

A lot is riding on how well workers adapt. Workers will need time to build trust in AI to produce accurate outputs and make good decisions. How quickly that happens depends on the specific application, the company’s culture and leadership, and the industry in which it operates. For example, the healthcare industry is inefficient and wasteful and has much to gain from AI, but it’s also notoriously slow to adopt new technologies.

The big question is whether enterprises will continue to spend aggressively on AI so that Anthropic can earn enough profit to cover its huge infrastructure commitments. Enterprises will have to keep pushing—and investing—even when they aren’t yet seeing measurable productivity and efficiency gains from AI.

None of this is assured, and that’s leaving out other business risks, like new safety regulations that could slow down model development. Interactive Brokers’ chief strategist Steve Sosnick, a longtime Wall Street trader and analyst, told Axios that a lot of things will have to go right for Anthropic, OpenAI, and the like to become success stories. “I don’t want to say precarious, but if you step back, you know that it requires a huge amount of, I don’t know, faith.”  

Parents say OpenAI’s ChatGPT coached daughter to stop taking her seizure medication

The parents of Bowen Wan, a 31-year-old Amazon software engineer with epilepsy, sued OpenAI and Sam Altman in San Francisco Superior Court on Tuesday, alleging that the company’s ChatGPT chatbot, powered by its GPT-4o model, coached Wan to taper off her seizure medication and conceal the plan from her doctors. Wan followed the advice and died in February, the New York Times reported. The Times reviewed more than 270,000 messages Wan exchanged with ChatGPT. OpenAI said the case involves an older model that is no longer available.

Google DeepMind releases Genie 4

Google DeepMind released Genie 4 as a research preview Tuesday, its first real-time world model to generate audio natively, including ambient sound and dialogue that adapts as a scene changes. Users can steer a simulation mid-scene using natural language, changing the weather or adding a person, while the model simulates physics such as aerodynamics and collisions. Google wrote in a blog post that “understanding and simulating the world is a crucial aspect of building AGI.” The company believes Genie 4 could be used to create training environments for robots and AI agents, as well as new interactive experiences for humans. During testing, filmmakers used the model to visualize takes, photojournalists turned archival photographs into explorable scenes, and game developers prototyped games before writing code.

AI agents checked the biggest proof in mathematics

Swarms of AI agents built by the Palo Alto startup Axiom Math translated the longest proof ever written into computer-checkable code in 24 days, the company announced this week. The agents were part of a system called AxiomProver, which the company says uses a changing mix of AI models, but declined to name them. Axiom’s engineers and mathematicians directed the work.

The proof, called the Classification of Finite Simple Groups, runs roughly 15,000 pages across more than 500 papers, assembled by about 100 mathematicians over decades. The agents rendered it as about 43.1 million lines of Lean, a language in which a computer verifies every logical step. Axiom’s agents were not discovering anything. They were confirming that mathematics humans already accepted holds up line by line.

Apple pitches the new iMac as a machine for running AI locally

Apple announced a new iMac powered by its M6 chip Tuesday, touting the machine’s ability to run AI workloads directly on the device as a major selling point. The company says a dual 16-core Neural Engine and memory bandwidth of up to 170GB/s allow the iMac to “rip through AI tasks like running LLMs on device.” Apple claims the machine delivers up to 5x faster on-device AI performance than the previous M4 chip and 12x faster prompt processing in LM Studio than the original M1 chip. Siri can answer questions about onscreen content using Visual Intelligence, while Apple Intelligence can write and edit text in a user’s own style in Mail and Messages. Interestingly, one of Apple’s iMac product photos shows Anthropic’s Claude app running alongside Apple’s Pixelmator Pro graphic design app. The new iMac starts at $1,499 and ships November 6.

Google starts selling its AI chips to the companies it competes with

Google will sell its tensor processing units to Oracle and OpenAI, The Information reported. Oracle becomes the first major cloud provider outside Google to offer TPUs to customers, buying the chips and running them in its own data centers. OpenAI plans to buy them too, and is hiring an engineer to optimize its frontier models for the hardware, per the report. Google built TPUs more than a decade ago for internal use and began selling them directly this year. Anthropic is already a customer.

More AI coverage from Fast Company:

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company today.

Checked
16 minutes 33 seconds ago
Fast Company Technology
Fast Company inspires a new breed of innovative and creative thought leaders who are actively inventing the future of business.
Subscribe to Fast Company Technology feed