That is really easy to use. I haven't tried any game design or anything like that for a few decades now; but in my 10 minutes playing around on it, I liked it a lot.
This has some very interesting privacy and security risks.
If the tech can do more complex frequency analysis, then couldn't it essentially be used as a microphone for a device that doesn't need permission.
It's a pretty well-known exploit that the CIA is capable of turning a lot of electronics with speakers into microphones. I imagine there is an entire classified backlog of things they can turn into microphones without the target's knowledge.
Idk though when I read it, it seems like it's literally an antenna attached to a can "resonator" is that electronics? It is I guess since it can carry an RF wave? Electronics I think of a chip or circuitry. I get it has to be some form of a circuit to work even as a monopole.
The article says it though: "...hung in his office behind his desk, and which contained an electronic device"
I think if you saw it on a bench hooked up to wires, it would be intuitive that it was a circuit. It's equivalent, but instead of being coupled via wires it's coupled via RF. I think it feels like there's no return path and that the circuit is open, but it's a real circuit with complicated/uncommon coupling to the power source.
A resonator is both a component in the circuit (the case is a cavity resonator) and the type of circuit this is. When illuminated (or hooked up to a power supply on the bench), it produces a sine wave, and holding all else equal the frequency is a function of the capacitance of our membrane capacitor. That membrane is flapping about due to sound, changing the distance between the plates of our capacitor and thus it's capacitance. So this shifts the frequency we're resonating at and encodes the audio into our output signal (frequency modulation).
So it's very similar to a standard LC resonator circuit you might make on a breadboard.
I'll leave you with another story of clever KGB sabotage. The KGB controlled facilities used to construct the US embassy in Moscow in 1979. They were able to extensively bug the building. They were also able to mix thousands of diodes into the concrete. This defeated NLJD (https://en.wikipedia.org/wiki/Nonlinear_junction_detector) based bug detection because they detected the diodes in every direction.
The CIA…plug a set of regular headphones into a microphone jack, open a recording application and speak into the headphone speaker, you don’t need a 3 letter agency for that physics open secret.
I got this clone Apple lightning to 3.5mm headphone jack adaptor for iPhone, my mind was kinda blown when I found out it just uses the lightning for power and inside the plug is a tiny bluetooth device that stream music over the 3.5mm jack. The original adaptor doesnt work like this as far as i have considered.
Wouldn't you need to rewire the headphones? Headphones use a 3-pin TRS whereas a 4-pin TRRS plug is used when you add a microphone. Regardless if the 4-pin is CTIA or OMTP, it's generally only going to get shorted to ground if a 3-pin TRS plug is plugged into a 4-pin TRRS socket, or if a 4-pin TRRS plug is plugged into a 3-pin TRS socket.
This is basic physics controlling the effect here, not electrical routing. Speakers are microphones by their very design. To make them work as a microphone, you merely speak into them with them plugged into an input jack that provides at minimum a line level electrical signal to be modified by wiggling the speaker cone/diaphragm back and forth.
Yes, but the computer doesn't have the firmware to "record" that signal from the speaker output pins. Thus, to record from the speakers acting like microphones, would require rewiring the headphone cable, for the vast majority of computing devices.
If you click "record" on your computer, there's no way to tell it to record signal from the speaker output channels, even if you write a custom low-level application directly making OS calls. The OS can't even do it, because it's not supported by the firmware.
"Yes, but the computer doesn't have the firmware to "record" that signal from the speaker output pins."
No, you plug directly into the microphone jack, that is what is providing your line level reference signal that gets changed by motion in the diaphragm. Zero rewiring required.
I am crap with physics but was going to say I think the last 50+ years of speaker development has been about making them less a microphone than they inherently are.
Dynamic loudspeakers and dynamic microphones are the same thing. They always have been the same.
They've got the knobs for the design variables turned in different directions, but they're still the same.
They even have the same frequency response whether they're being used as speakers or microphones at the moment.
Which brings up a valid way to measure the response of a microphone's design:
Use two of them. One as a speaker, and the other as a microphone. Play measurement-sounds out of one, and record the results on the other. Plot it out.
The deviations are magnified, but eliminating that magnification is just a math problem -- not an instrumentation problem. :)
They transmit sound. Anything able to detect the vibrations make it a microphone. Not sure how a speaker gets around that because it’s job is to vibrate.
There is something better. The little sensor that maintains the distance between the spinning platter and the armature is sensitive enough to be a reasonable microphone. But it is inside a heavy metal box (the HDD) so you do need to shout at it.
>> They tapped into the feedback system that helps control the position of the read head above the magnetic disk. When the head is buffeted by sound waves, the vibrations are reflected in the voltage signal produced by the drive’s position sensors. By reading this signal, Fu and his colleagues were able to make high-quality recordings of people speaking near the drive.
The NSA could turn on your flip phones mic thirty years ago without you knowing, I don’t think they needed to do all that fancy stuff with hard drives. That’s just research that they funded to cover up the fact that they owned every computing device on the planet for a while.
I don't think that's realistic. If you're looking at the acceleration sound waves cause against a phone's accelerometer, that's likely far below the sensitivity of the sensor- phones are too massive relative to the force of sound waves from speaking. F=ma, so the acceleration you're looking at is the force of the soundwave (tiny) divided by the phone's mass (relatively large). The only reason this kind of works is because you're putting the phone on an object that's mechanically vibrating. I suppose it would work in certain situations like putting the phone on top of a large speaker, but you'd never get the resolution to decipher audio from sound waves alone for a phone sitting on a desk or in a pocket
I doubt the sampling rate is anywhere near what you would need to make out dialog in a sound recording. You might be able to tell who is speaking though if you had a voice profile.
I feel like lots of people here are just commenting on the headline.
This isn't about the local models you're running on your old gaming rig, or the tesla p40 rig you build for local llm's.
This is about code leveraging the local resources where the code is running for it's AI needs. Rather than making an API call to an external AI service, the code leverages the AI capabilities built into the hardware it runs on. With modern Apple, Intel, and AMD silicon all shipping dedicated AI acceleration, this is the where IMO the focus should be heading.
How many Flops or whatever can your phone do?
I bet it's enough to paint the walls of your living room, or draw a pretty good pelican on a bike.
And this is exactly what the LLM provider industry is fighting tooth-and-nail. It’s not only because it doesn’t directly contribute to their bottom line, it also directly opposes the idea that LLMs are going to replace entire workers rather than enhance the abilities of individual workers. What we’re headed towards would have been a killer product and probably still shifted a bunch of capital to the bazillionaires had these companies set more realistic goals rather than banking that they’d be the ones that won the war that “changed everythingTM”.
As long as Apple and Google put reasonable AI capabilities on device, then software engineers will use those capabilities when it makes sense (the article gives lots of good examples of capabilities that make sense to run locally). As the author notes, it's cheaper and more reliable to run these things locally.
That also doesn't preclude LLM services from being massively successful, they'll just have to justify the pricing and complexity that comes with their adoption, just like any other product.
> That also doesn't preclude LLM services from being massively successful, they'll just have to justify the pricing and complexity that comes with their adoption, just like any other product.
What is completely different from every other product is how much they’re spending, and how much they’re obligating themselves to spend going forward. I think there’s a very good chance that the existing providers could be miles underwater coming out of this. Even if the business is not the everything to everybody that they’re banking on it being, they still owe all of that money back to the people they borrowed it from, and they will be a lot less likely to float them cash to get them back to a normal operating mode if they burned the last ocean of cash promising the universe and winding up with “oh yeah, that’s pretty useful sometimes.”
Yeah that's a good callout for sure, the spending here is nuts so agree that it's not "just another business that has to price itself right to be competitive".
I guess if the time horizons is long, like 20 years, then maybe the spending, as it begins to amortize, gets more in line?
I was thinking that a comparison could be to cloud providers, each of which had to spend a lot of money to build out datacenter before making money. Difference there is AWS proved the product first, so when Microsoft and Google came along, they knew it would work and be profitable. With AI, nobody has proven it will work and be profitable, they're all competing for that at the same time which is a potentially dangerous mix for the reasons you cited.
The only way that this even vaguely works, best I can tell, would be on that decade-or-two timeline, but therein lies the problem: all this money getting pumped into data centers right now is going to produce data centers that are running old, inefficient, slow GPUs by 5-years-from-now standards. And GPUs are by far the most expensive part of these data centers… having the buildings is barely an asset. We’re investing all the money in right now’s technology in one of the fastest moving hardware segments and for some inexplicable reason, think that will lead to a sustainable advantage. What’s to stop someone 5 years from now, waiting for the dust to settle, then spending way less money for more compute and just mopping the floor with everybody in this sector… and that’s (unreasonably, IMO) assuming that local applications won’t become good enough to take too large a bite from their business before that.
And look at the difference in spending between their building out general-purpose-computing cloud data centers that even then, had potential use cases if the business failed. What are they going to do… start a massive, extremely expensive pre-rendered online gaming service? Only render Disney movies?
These datacenters are already running old, inefficient, slow GPUs from five years ago in addition to newly released cards, because anything newer than that is extremely bottlenecked and they need all the compute they can get. Why should it be any different in five years' time? Even nVidia is rumored to be about to bring back the RTX 3060 which is an Ampere architecture card that got released around 2021. It's just fine.
If those data centers were good enough, they’d save themselves a few billion dollars and just do more of the same, wouldn’t they? Many current video games struggle on the 3060— it’s like 10 times slower for interference than a 4090 even. They’re reintroducing it because their upstream business of selling brand new insanely expensive GPUs required for every new data center is making it impossible for people to buy GPUs for their home computers. It says nothing about data-center-class GPUs except that every company currently has a burning desire to only have the latest and greatest GPUs.
The new GPUs are a lot better than the old ones, to be sure. They're also a whole lot harder to get ahold of in quantity. That's no different than the reason for officially reintroducing the 3060.
Yeah, conceptually this isn't all that different from new VM SKUs coming out in clouds. The costs and rate of change for AI hardware may be higher, and perhaps enough higher to mess up the math, but conceptually its a model that has been proven to work.
like by selling it at a loss to build dependencies and then jacking the price up year after year by whatever amount is just below the cost of removing the dependency
> ...it also directly opposes the idea that LLMs are going to replace entire workers rather than enhance the abilities of individual workers.
Which also, as I feel the need to remind everyone every time it comes up, has not yet once been actually shown to be a workable strategy. For any worker in any industry.
And to be clear, I'm talking about a worker, sitting in a chair, replaced with an agent, sitting in... a server, I guess, where nothing else about the org has to be changed. That's what's being advertised and sold, and it has never to my knowledge actually happened.
If their product is "access to a big model running on a really big computer" (if we can count 'multiple data-centers' as a single enormous distributed computer), then the product "small, accessible device that everyone has" risks killing their cash cow.
Ironically enough, the first company to really focus on "an LLM in every phone" will have a good shot at actually being the ones that "changed everythingTM", in the way Microsoft changed the world from IBM mainframes to PCs, or Apple made smartphones a thing.
As an aside, the mainframe industry was profitable for decades before PCs took over. It’s not like they spent a zillion dollars ramping up at the same time.
Actually you can do way more things than that. We have optimized it to process 2TB of high def videos on a M5 MBP in under 24 hours, including everything such as speech understanding, face recog, LLM and VLM. Super fun.
If Steve Jobs was alive Apple would have already demoed this as a new line of Macs with open weight models pre-installed with hooks into all of their existing content creation software.
And he would have the audience believing all the demos were running through third party AI providers, until at the last moment explaining “actually all of that ran on device with no connection to any external services.”
It was a famously hard task. It was an ingenious idea for an unexpected task that falls outside of the bounds of predictable normal input but is still readily comprehended by the public.
Unfortunately, as soon as it's a famously hard task trainers know they need to succeed at it and it loses a lot of the power to detect correctness.
Maybe this is an example of training overfit. But it won't be too long before local models chew through the "famously hard tasks". Except possibly ARC-AGI. That's one benchmark that is still developing with capabilities. And every time a new ARC-AGI benchmark is released it make the SOTA LLMs look pathetic. Because there is very little understanding or transferability with LLMs. But in terms of benchmark-able micro tasks, the local LLMs are improving.
I just did something exactly like this. I have a self-hosted personal dashboard and one of the APIs I'm reading gives slightly too verbose of an output. So I added a feature to summarize the text using Qwen 3.5 2B which happily runs on a CPU. I've never clocked the tokens per second because I only generate like 100 tokens an hour in a very narrow domain of knowledge and speed isn't critical.
A phone makes a very crappy AI inference rig. It's battery powered and can't even really run at 100% utilization on an ongoing basis due to how challenging the thermals are.
at the moment yes. The one possible silver lining with all of the current hardware crunch is that it _should_ force some hardware advancements. The last couple years hardware has been kinda boring. My m1max is still zippy as all hell and doesn't really need to be upgraded, unless I am committing to local AI inference.
I kinda assume phones are going to be battery powered for the foreseeable future. "Gaming" phones with better cooling do exist, but they are a tiny niche. Most local AI users will want to serve their inference needs through a very different kind of system.
Yes, but the battery tech itself is improving. We're already seeing new phones approach 8000 mAH internal batteries, which is large enough that you can splurge on compute and still have some left over at the end of the day.
I'm very curious what kind of hardware advancements you're imagining. Because we're already kind of near a physical wall regarding heat dissipation on phones.
I mean hey, maybe foundational physics will surprise the world with a radical breakthrough that disappears heat into a black hole or something, but I sure wouldn't hold my breath
eGPU cradles, presumably, for people with intense local model execution requirements until it can be made to work in the device? This is exactly like the POS dongles Square had until tap to pay was more widespread?
I was writing just about this last week for fun: AI + hardware team-up to build localized AI with specialized functions to your organization. Ex. Adoble Studio AI in an on premise Box, made by Apple and powered by something like Cohere with privacy:
Cheap mini pc on an isolated vlan. Running a cloudflared tunnel or reverse proxy to a vpc on digitalocean, maybe moving to Hetzner soon, or in addition to for failover/ha.
All containers. Some just docker/podman, some one k8s cluster. Mainly it's just for fun. Except the cloud and local backups for our phones/Gcloud, which my wife will get really mad if it doesn't work.
clean setup! I have a raspberry pi but haven't configured it yet.
a few questions if you don't mind:
1. how do you manage deployments? Is it some sort of sh script or CI/CD pipeline?
2. why do you choose k8s for some services? just for learning?
Gitops. I use janky infrastructure as code to manage the services running on any stateful machines, and mainly just stuff I through together.
I run k8s because everyone runs k8s now, and it's good to have a local environment to play with.
With the majority of popular music decided by a handful of people in LA, Nashville, or New York, let me ask one question: is this actually a bad thing?
Hear me out. Most of what's on the radio could have been made by AI already and no one would've noticed the difference.
To be clear, I'm not talking about legitimate artists doing something original or authentic. I'm talking about the execs who find performers to sing and dance over their perfectly manufactured hit single. Songs made by people like Max Martin, who aren't trying to express anything beyond their knowledge of which combination of notes has the highest ROI. No disrespect to Max, he's incredibly talented at what he does. But now the execs have the data, and they don't need the Max Martins, Diane Warrens, or Carole Kings anymore. They can plug in the numbers and out comes a perfect song for their next artist.
So let's embrace the new AI pop. Let it dethrone the kings who've shaped the sound of pop culture for too long.
Real art always seems to find its fans eventually, and I don't think AI will stop that. Yet. When a model writes a song that lingers the way "Linger" does, maybe it will. But at that point, if the music really is that good, does it matter?
> Real art always seems to find its fans eventually, and I don't think AI will stop that.
That's what's happening though. Someone is writing a prompt, generating some slop and then directing bots to collect money that will never make it to real artists.
If they were getting too small of a slice before, now it's even smaller.
> But at that point, if the music really is that good, does it matter?
We're very far from "that point". The point we're at right now, which we need to tackle is the one in which we're drowning in slop.
I'm not personally against using AI tools to write code or help someone generate music, but right now it just impedes, devalues and deplatforms the real stuff to make money from a system that hasn't caught up.
Two things. One, surely I'm not the only one who knew this data was being stored? Two, calling it a "leak" feels like a stretch when the data was publicly accessible by design from the start.
Yes, some users probably didn't realize their edits to public pages were saved publicly, and that's a legitimate UX complaint. But some of the responsibility has to sit with the user. Otherwise we'd be running daily headlines about Meta "leaking" user data to every advertiser with a checkbook.
So many people are victims of what I like to call the "Field of Dreams Fallacy."
Also known as: If you build it, they will come.
In the real world, it doesn't matter in the slightest if you build the best product in the history of the universe. If you don't have the proper marketing and sales pipelines, you will lose to the product that does.
There was a great article on here a while back about VHS and Betamax. While Betamax was better by nearly every metric, it lost.
Same for HD-DVD against BluRay. And for so many other great products that have died on the vine.
I think this is actually a bigger problem with society as a whole than people notice. The majority of people think that an idea alone is as valuable as a business. People regularly tell themselves that if they would have come up with X, Y, or Z, then they'd be rich and successful. When in reality, the product or idea doesn't equate to success in the slightest.
It's the same thing that I'm sure a lot of you in tech see, where people say "Can you make me an app?" or "We should start a tech company that does this one thing better than the other guy." And yet almost every single time I explain to them that there are 4,000,000 versions of their app already, and that it's still a business that requires significant effort, they act like it's my fault for not helping them or not believing in their idea.
I've let millions of better ideas fade from memory without a second thought. Because I've learned that operating a successful business is an entirely different world from building a cool thing. The idea is the easy part. Everything after it is the actual work.
"The idea is the easy part" is something I knew deep down but was too afraid to admit to myself. It's easier to keep building than to confront that everything after it is the actual work. Took me years and a few dead products to really accept that.
I feel like Im in the process of swallowing that pill.
> There was a great article on here a while back about VHS and Betamax. While Betamax was better by nearly every metric, it lost.
There is some nuance here.
Manufacturers didn't know if people preferred shorter, higher quality (Beta) or longer, less quality (VHS). That's partly why there were two formats.
Most people like to say VHS "won" but what it really won was the consumer market. Beta won the professional TV news market because it turns out news stations had a high demand for short, high quality video storage.
I point this out only to say that winning isn't a one dimensional/binary outcome. You can "lose" in one market but still be very profitable in another market.
>There was a great article on here a while back about VHS and Betamax. While Betamax was better by nearly every metric, it lost.
Not every measure. It could hold like half the recording as a VHS.
The “killer app” for video tapes was being able to record live TV when you weren’t home with timers on the players. VHS could record more without any intervention.
It also meant most movies could be watched all the way through without interruption.
People didn’t care about the ever so slightly better quality of Betamax. They cared about getting their moneys worth and not being interrupted.
What mattered was every other company got together and agreed on a format. The size of their market caused hollywood to release more movies in that format which made more people buy. The super long mode became more important later.
It's 2026, you should not be using command prompt (or batch.) In powershell ls is a built in alias to get-childitem and has been for years, and in recent versions of windows you'd have to go out of your way to get a command prompt (you would have to open a powershell terminal and then run cmd.)
On one our linux machine filesystem became strange, probably because somebody mistyped `ls /bin` as `ln /bin`. I think docs say hardlinking folders is impossible or maybe /bin was a symlink.
