Jamming on the parking brake, when going 90, down the highway, is a bad idea.

I sometimes miss a turn, or don't plan well enough to be in the correct lane, when I arrive at the intersection.

What I do, is go "D'oh!", continue to the next intersection, then either make a U-turn (if legal), or turn onto a side street, with the intention of recovering my intended direction.

What I often see people in the same situation do, is jam on the accelerator, swerve across six lanes of traffic, and screech into their turn.

That may get them where they are going, but it also has a very real chance of earning them a ticket or a stay in hospital.

My way takes a bit longer, but no ticket, no accident.

Except when they get in an accident trying to force it. Or when they were too distracted unnecessarily overtaking someone near their turn.

Damn straight.

Remember that every time we query an LLM, we're giving it ammo.

It won't take long for LLMs to have very intimate dossiers on every user, and I'm wondering what kinds of firewalls will be in place to keep one agent from accessing dossiers held by other agents.

Kompromat people must be having wet dreams over this.

BigTech already has your next bowel movement dialled in.

Someone would have noticed if all the phones on their network started streaming audio whenever a conversation happened.

It would be really expensive to send, transcribe and then analyze every single human on earth. Even if you were able to do it for insanely cheap ($0.02/hr) every device is gonna be sending hours of talking per day. Then you have to somehow identify "who" is talking because TV and strangers and everything else is getting sent, so you would need specific transcribers trained for each human that can identify not just that the word "coca-cola" was said, but that it was said by a specific person.

So yeah if you managed to train specific transcribers that can identify their unique users output and then you were willing to spend the ~0.10 per person to transcribe all the audio they produce for the day you could potentially listen to and then run some kind of processing over what they say. I suppose it is possible but I don't think it would be worth it.

> Google agreed to pay $68m to settle a lawsuit claiming that its voice-activated assistant spied inappropriately on smartphone users, violating their privacy.

Apple as well https://www.theguardian.com/technology/2025/jan/03/apple-sir...

I keep seeing folks float this as some admission of wrongdoing but it is not.

The money is the admission of guilt in modern parlance.

While not an "admission of wrongdoing," it points to some non-zero merit in the plaintiff's case.

It absolutely is.

If they knew without a doubt their equipment (that they produce) doesn't eavesdrop, then why would they be concerned about "risk [...] and uncertainty of litigation"?

Also people already believe google (and every other company) eavesdrops on them, going to trail and winning the case people would not change that.

Again: If their products did not eavesdrop, precisely what risks and uncertainty are they afraid of?

(1) Alphabet admits wrongdoing, but gets an innocent verdict

(2) Alphabet receives a verdict of wrongdoing, but denies it

and the parent using either to claim lack of

> some admission of wrongdoing

The court's designed to settle disputes more than render verdicts.

It's a private, civil case that settled. To not deny wrongdoing (even if guilty) would be insanely rare.

Only if you use a very narrow criteria that a verdict was reached. However, that's impractical as 95% of civil cases resolve without a trial verdict.

Compare this to someone who got the case dismissed 6 years ago and didn't pay out tens of millions of real dollars to settle. It's not a verdict, but it's dishonest to say the plaintiff's case had zero merit of wrongdoing based on the settlement and survival of the plaintiff's case.

You don't have to stream the audio. You can transcribe it locally. And it doesn't have to be 100% accurate. As for user identify, people have mentioned it on their phones which almost always have a one-to-one relationship between user and phone, and their smart devices, which are designed to do this sort of distinguishing.

If this really is something that is happening, I am just very surprised that there is no hard evidence of it.

With their assumptions, you can log the entire globe for $1.6 billion/day (= $0.02/hr * 16 awake hours * 5 billion unique smartphone users). This is the upper end.

At one point I had the misfortune to be the target audience for a particular stomach churning ear wax removal add.

I felt that suffering shared is suffering halved, so decided to test this in a park with 2 friends. They pulled out their phones (an Android and a IPhone) and I proceeded to talk about ear wax removal loudly over them.

Sure enough, a day later one of them calls me up, aghast, annoyed and repelled by the add which came up.

This was years ago, and in the UK, so the add may no longer play.

However, more recently I saw an ad for a reusable ear cleaner. (I have no idea why I am plagued by these ads. My ears are fortunately fine. That said, if life gives you lemons)

So isn’t it possible that your friend had the same misfortune? I assume you were similar ages, same gender, same rough geolocation, likely similar interests. It wouldn’t be surprising that you’d both see the same targeted ad campaign.

The only reason I was served the ad was because I had an ear infection months before.

Plus this was during covid. So this was the smallest group size permissible and no one else around for miles.

The big AI companies have not really demonstrated any interest in ethic or morality. Which means anything they can use against someone will eventually be used against them.

> The big AI companies have not really demonstrated any interest in ethic or morality.

You're right, but it tracks that the boosters are on board. The previous generation of golden child tech giants weren't interested in ethics or morality either.

One might be mislead by the fact people at those companies did engage in topics of morality, but it was ragebait wedge issues and largely orthogonal to their employers' business. The executive suite couldn't have designed a better distraction to make them overlook the unscrupulous work they were getting paid to do.

The CEOs of pets.com or Beanz weren't creating dystopian panopticons. So they may or may not have had moral or ethical failings but they also weren't gleefully buildings a torment nexus. The blast radius of their failures was less damaging to civilized society much more limited than the eventual implosion of the AI bubble.

And now that they themselves are targeted, suddenly they understand why it's a bad thing "to give LLMs ammo"...

Perhaps there is a lesson in empathy to learn? And to start to realize the real impact all this "tech" has on society?

People like Simon Wilinson which seem to have a hard time realizing why most people despise AI will perhaps start to understand that too, with such scenarios, who knows

The community is often very selfish and opportunist. I learned that the role of engineers in society is to build tools for others to live their lives better; we provide the substrate on which culture and civilization take place. We should take more responsibility for it and take care of it better, and do far more soul-seeking.

If the author had configured and launched the AI agent himself we would think it was a funny story of someone misusing a tool.

The author notes in the article that he wants to see the `soul.md` file, probably because if the agent was configured to publish malicious blog posts then he wouldn't really have an issue with the agent, but with the person who created it.

Be careful what you imply.

It's all bad, to me. I tend to hang with a lot of folks that have suffered quite a bit of harm, from many places. I'm keenly aware of the downsides, and it has been the case for far longer than AI was a broken rubber on the drug store shelf.

Just saying, what you're describing is entirely unsurprising.

This is a front-page link on HackerNews. It's going to be referenced in the future.

I thought that they handled it quite well, and that they have an eye for their legacy.

In this case, the bot self-identifies as a bot. I am afraid that won't be the case, all the time.

There's the yellow composite plug, a 12V input, and a small bit of wire to be cut to rotate image 180 degrees, at the other end of a 30ft cable from the camera. The composite goes into the existing infotainment. There would be a wire from shifter to infotainment that switches the display to external composite video when the gear lever is in reverse. I think it even came with a miniature hole saw in size of the camera module.

$10 and one afternoon later, I could have upgraded a dumb car to have one, complete with auto switch to backup on reverse. No software hacking needed. It's fundamentally an extremely simple thing.

Much as I despise them, I'm not so sure that would be the case. I seem to remember folks saying the same about the Taliban, and the cartels have a lot more money and high-tech kit, than the Taliban.

Asymmetric warfare is a tough gig, on all sides.

* A conventional military war, on a battlefield: Neither Saddam Hussein's military nor the cartels nor the Taliban would last long against the US.

* An unconventional insurgency: The Iraqis quickly turned to this approach and it worked very well for them, as it did for the Taliban. The Taliban won, and the Iraqi insurgency almost drove the US out of Iraq and was eventually co-opted.

The cartels of course would choose the latter. They, the Taliban, etc. are not suicidal.

The US decided to leave because staying was not politically popular, and left. They were not beaten by the Taliban, they were beaten by the political climate at home.

If someone is actively kicking your ass, then they decide that you aren't worth the effort to keep hurting and decide to walk away, that doesn't mean you "won" the fight even if you get what you want afterwards.

It's also a perfectly common, expected way to win a war: First, wars always end with political solutions. The most well known principle of warfare is that it is 'politics conducted by other means' (i.e., by violence rather than by law or diplomacy). If there is no political solution, the war never ends. That's why the US didn't win the war in Afghanistan after decades - they couldn't create a stable political solution because they were unable to impose one on the Taliban, who in the end imposed one on the US and its allies.

Victory by outlasting enemy resources, including political will, is fundamental to warfare; wars end when resources to fight (for the political outcome) run out, but few end in total kinetic destruction of those resources - someone runs out of money or political will. It's also the explicit strategy of insurgencies. Enemies of the US know it very well and have used it for generations - that is how North Vietnam won, for example. When the Soviets invaded Afghanistan, the Afghans famously told them, 'you have the clocks (the technology), we have the time'.

Your parents get frustrated and leave. You now take a cookie from the jar.

You have a cookie, but that doesn't mean you won the argument.

I mean, some sort of cartels would bounce back after any "war on drugs" because supply and demand, but the people running them aren't hankering for martyrdom or glory over consolidating their territory and accumulating.

There isn't a real analogy there because cartel leaders have no official state support anywhere, let alone in a bordering nuclear power, but even if they did, it hardly seems reassuring from their perspective to know the drug trade will outlive them after they all get killed. It's different when you're deeply religious and believe what you're doing is worth dying for and the larger arc of history is more important than your own life and wellbeing. I don't think drug lords think that way.

I didn’t mean to start this giant thread about Mexican Cartels but here we are. Most think it’s just an isolated problem. Others know it’s more widespread. I simply stated that these murderous thugs are out there in full force with technology and armored vehicles. If provoked, they would lash out. It’s ridiculous because of course going up against the US is a losing proposition but each “generation” of cartel leader thinks they can somehow manage it.

I'm sure that image nerds would poke holes in it, but it seems to work pretty much exactly the way it does IRL.

The noise at high ISO is where it can get specific. Some manufacturers make cameras that actually do really well, at high ISO, and high shutter speed. This seems to reproduce a consumer DSLR.

Even on old, entry-level APS-C cameras, ISO1600 is normally very usable. What is rendered here at ISO1600 feels more like the "get the picture at any cost" levels of ISO, which on those limited cameras would be something like ISO6400+.

Heck, the original pictures (there is one for each aperture setting) are taken at ISO640 (Canon EOS 5D MarkII at 67mm)!

(Granted, many are too allergic to noise and end up missing a picture instead of just taking the noisy one which is a shame, but that's another story entirely.)

The amount and appearance of noise also heavily depends on whether you're looking at a RAW image before noise processing or a cooked JPEG. Noise reduction is really good these days but you might be surprised by what files from even a modern camera look like before any processing.

That said, I do think the simulation here exaggerates the effect of noise for clarity. (It also appears to be about six years old.)

Yes, this simulation exaggerates a lot. Either that, or contains a tiny crop of a larger image.

I do feel (image nerding now) that its shutter/ISO visual for showing the image over/under-exposed is not quite correct. It appears they show incorrect exposure by taking the "correct" image and blend (multiply) with either white or blend with black (on the other end of the exposure spectrum) to produce the resulting image.

I suppose I am expecting something more like "levels" that pushes all the pixels to white (or black) until they are forced to clip. (But maybe I am too trained in photo-editing tools and expect the film to behave in the same way.)

Lately I've been researching cameras for astronomy, especially for deep-sky objects (DSOs) like nebulae that require hours of exposure time. The marketing material for these cameras go into a lot of detail: quantum efficiency (the percent chance that a photon converts into an electron), dark noise at different temperatures (fractions of electrons per second), readout noise (usually around 1 electron), and well depth (usually around 10k electrons). Compared to general photography, the astro community much more motivated to explain and keep track of all the sources of noise. Random product example: https://www.zwoastro.com/product/asi585mc-mm-pro/

Very limited camera choices, though.

I'm interested to see how the roll turns out - gave it for development the other day, had a good laugh with the employees though.

I now have a mnemonic for it: Blor - a (somewhat) portmanteau of Blur and low. So low aperture = blur.

Edit for clarification: I mean low number (2 vs 32) = blur

Unfortunately the lower number actually means bigger aperture.

With my mnemonic, I say low *number = blur

I should have been more specific

Denominator, not numerator. That's why larger number = smaller aperture.

But photographers generally just say "f2", meaning an aperture value of two set on the dial of the camera/lens. It's one stop faster (twice as much light) as f/2.8. It'll give you a relatively shallow depth of field, but not as shallow as e.g. f/1.4.

The smaller, i.e. the closest to an ideal pinhole camera, the wider the depth of field is. A an ideal pinhole camera has infinite depth of field.

Unfortunately the aperture f numbers are the wrong way round; larger numbers correspond to smaller diameters.

I have a friend that went to school in Buffalo, NY. That’s a city that experiences “lake effect” snow, during the winter.

He says all the sidewalks are basically “snow gorges,” but the roads clear quickly, and everyone knows how to dress for the cold.

He tells me a story about visiting northern Quebec, one summer, and seeing houses with a second front door, set on the second floor, and was told they were “snow doors,” for deep winter, so folks can get out, when the snow gets deep.

We tend to adapt well.

He said they tended to build “two of everything,” which is why there’s so many sets of two.

If one craps out, the second can be used in its place, or scavenged for parts.

https://en.wikipedia.org/wiki/Unified_Payments_Interface

It sounds a lot like what they're discussing.

I was just commenting on how good, widespread it is and no it is not the doing of the current Govt. It just gained traction during a massive f up of the current Govt.

You can buy vegetables at a farmer's market, and just scan a QR code.