How can price discovery work if everyone’s price is different? What downward pressure would even exist on surveillance-derived prices besides resale black markets? Societally, do we really want to further monetize the expansion of the mass surveillance state?
> In July, Meta announced that it would push an update to detect when an LED has been “physically tampered with or destroyed” and stop people from being able to film
Would love to know how they plan to do this, sounds like a patch that is bypassed with a $0.01 resistor.
Worst thing that's ever happened to me is having processes randomly killed by the OOM killer. Could be unpleasant if you're writing an email or document that suddenly disappears because your app got killed.
The OOM killer usually kills the process that's using the most memory, which is the reason the system is OOM. Sometimes this can be adjusted. When I'm OOM it's almost always because some runaway loop used 50GB and it makes no sense to kill any other process.
If it's killing your email editor then your email editor is bloated or you just need more RAM - or swap.
It highly depends on your specific workload, but freeing up RAM by paging out rarely or downright unused pages to swap can definitely give a performance advantage.
In consequence, I always give swap. Even on the off-chance it ends up unused, disk space is orders of magnitude cheaper than RAM, so I don't mind sacrificing it for potential gains.
I feel a similar fate will befall engineers too. Your predictions anre quite interesting from that perspective.
Markets defined entirely by law have distorted our collective understanding of what can actually be built with the knowledge our species has accumulated thus far. How will traditional shields that have protected capital accumulation in tech to survive in a world where governments now realize control of technology is a national priority? Especially as we see its impact on modern warfare, and that such conflict looks like it’s only escalating over time.
Mathematicians appear to me (as an outsider) to exist in a field without such distortions, and I think offer engineers a preview of what’s to come. I certainly have completely ceased sharing original ideas online at this point.
Is the fact that Nvidia is also massively financing the closed-models companies, which stand to lose when open model project(s) take off, a conflict of interest here? I can’t tell if this is nvidia hedging or planning to embrace & extinguish.
Any model ran takes compute. Everybody, datacenters and local need compute.
And I don't see a future where datacenters lose much to open models. I've ran local server rooms before clouds were a thing. Cooling and power is a very expensive pain in the ass. You're not dropping a rack of local models in most buildings without breaking the local power grid.
What about when the open models get to the point that you don’t need these fancy GPUs? It seems inevitable this will happen over time, we see cracks right now within even just GPT w/ expert streaming and flash attention. ASICs are coming soon too.
AWS and Azure would love to cut out OpenAI and Anthropic and sell inference on open models directly to businesses with no cut to share, just as AWS serves other open source software.
Open models are great for everyone except anthropic and OpenAI.
It frustrates me when people call them license plate readers. I guess you need to call them something, but they are general-purpose internet connected cameras. They will do whatever their firmware tells them to do, and could be reprogrammed at any time by anyone with access. No one expected doorbell cameras to join a mass surveillance network, but later the manufacturers added that feature. Why do we treat these cameras like they can do only one thing?
> Why do we treat these cameras like they can do only one thing?
Because its useful messaging to disarm the populace. "License plate reader" sounds less alarming than "Internet connected camera" or "Mass surveillance device."
Public attitudes vary around a "license plate reader" but will be more uniformly opposed to "Widespread, automated government mass surveillance tracking your every move via camera in real time"
Security cameras have been internet connected for a lot longer than we had this kind of system where they all have computer vision running and tagging everything in a massive database.
They were always internet connected and mass surveillance. It's the addition of search that changed things.
I'm holding a more powerful surveillance device than an Internet connected pole cam which we normally term a "smartphone". Its ability to spy on folks is widely known and almost no one seems to care enough to change their behavior. You'd be surprised what people will or won't care about. Once Trump is out of office, the folks currently obsessed with privacy will move on to the next thing...
They don't need to. They know where you've been, where you're going, and what your intentions are. It's a far more intimate and effective surveillance tool. As a bonus, it's also a propaganda vehicle and an implement of mass psychological manipulation.
Oh yeah sure. Parallel construction doesn't exist. Government surveillance is completely benign as long as they can't bring a criminal case with the information, right? I mean, we have rules now, so it's all good.
I agree parallel construction doesn't seem to be ordinary practice. But AFAIU the term originates from post-9/11 NSA warrantless surveillance--both mass and targeted--where information was often systematically "leaked" to the DoJ and FBI. Until recently states didn't have large agencies dedicated to collecting evidence that is patently illegal to use in criminal cases, and generally in a deeply grey area generally, which they then pinky promise not to share.
But "anonymous sources" and similar evidence laundering has been in a thing in domestic law enforcement since SCOTUS started imposing prophylactic procedural protections in the 1960s. It was just ad hoc. But now states like Texas and California have built their own intelligence agencies which are beginning to push the boundaries of both mass and targeted surveillance similar to the NSA, and through ubiquitous digital communications and third-parties like Flock, police departments have also organically grown their own whisper networks that allows them to systematically and efficiently do what was ad hoc before.
That said, unlike the post-9/11 NSA parallel construction which was knowing and deliberate and something of an open secret at both the NSA and parts of Federal law enforcement, I think state law enforcement personnel believe they're well within the law. And interestingly I was recently speaking with an FBI financial crimes investigator about this stuff and he had never even heard of the term parallel construction. Building durable parallel construction processes requires maintaining plausible deniability, and the best way to do that is to keep most law enforcement personnel and, preferably, all district attorneys out of the loop and ignorant to what's going on. That happens by high-level officials quietly tasking a handful of mid-level people to launder the evidence. And in that sense parallel construction will never be ordinary practice the way people assume as the vast majority of law enforcement personnel will never see it happening, or at least recognize it for what it is, except for when they do it themselves in an ad hoc manner.
Seems to me that as long as parallel construction doesn't lead to long prison sentences for the perpetrators, you should assume it's ordinary practice. Police and other surveillance authorities have zero legal reason to take their hand out of the cookie jar, and every reason to keep this out of public knowledge.
That's because that lookup is especially broad. And I'd guess the supreme court justices are worried about getting caught in one. For anything else, like downloading one person's videos, you don't need a warrant.
> No one expected doorbell cameras to join a mass surveillance network, but later the manufacturers added that feature.
Plenty of people did? That was obviously one of the end goals for cloud connected devices pushing recordings to a remote side for processing and storage. Especially when stuff like facial recognition was implemented server side.
I would be highly surprised if this was not pitched in the first internal product meeting.
While I give some grace to the regular population on this topic - anyone posting on HN should have trivially seen this outcome as not only likely, but nearly written in stone. Even if all your imagination captured was “secret NSA warrant” it would be one of the first risks you’d imagine for such a product.
It was obvious that this was the final outcome even back when Facebook first implemented tagging friends in photos before image recognition got cheap and good enough to go back and time and correlate the social graphs of everyone.
I don’t doubt it was intended by the manufacturer, but do you think the average customer knows this? Every normal person I’ve ever told this to, that their Ring cameras can also be looked through by the police, has looked at me like I’m crazy. I’m sure some people would buy a Flock camera for their porch, but I’d bet many others wouldn’t if they truly knew what they were signing up for.
The Super Bowl commercial that Ring put out about finding lost dogs spooked people for a reason!
> Plenty of people did? That was obviously one of the end goals for cloud connected devices pushing recordings to a remote side for processing and storage. Especially when stuff like facial recognition was implemented server side.
There is a huge difference between storing and processing recordings for a customer's own use, and aggregating them from multiple sources for organized mass surveillance. With that said, most corporations do not have the principles to resist the latter.
This amounts to an argument against ever using general purpose computing devices to accomplish anything.
You could say the same about literally any privately operated device. ‘I don’t know why they call these things cell towers. They do whatever they are programmed to and can be reprogrammed by anyone with access’
Heck the same applies to non electronic things too ‘why do they call it a house? It’s a general purpose structure and could be turned into a shop or a factory or a crack den at any time by anyone with access’
As others have noted, I believe the brightline is less “general purpose computer…” and more “…that is remotely accessible & modifiable effectively anywhere in the world”. I worry not just about OEMs and cops, but also criminals, foreign intelligence agencies, etc. Honestly potentially anyone at this point, now that folks have figured out you can use LLMs to do analysis.
This is a sound foundation for concern by itself, don’t you think?
I think that’s a very different concern - your take is more ‘governments shouldn’t deploy these things because they are incompetent’ whereas the original concern seemed to be ‘governments shouldn’t deploy these things because they are malicious’.
Neither philosophy seems to have a good limiting principle for what they mean governments should be allowed to do to efficiently enforce traffic law
Don’t get me wrong, governments putting these things up for one purpose and then using them for another (my read on your use of “malicious”) is a major concern I have.
It is broader than that though, and is the distinction I sought to capture with “anyone with access” in my original post. They are connected to the internet, and to a cloud no less! Anyone could be the NSA now, from anywhere in the world. I bet we’ll see these used to collect marketing data too.
Perhaps a limiting principle can be that we simply do not use ALPRs or other surveillance tech, or at least that we don’t connect them to the internet and require a warrant to access and use their data? Personally, I’d just prefer cops enforce traffic law the old fashioned way.
The old fashioned way cops enforce traffic law is, typically: unevenly, ineffectively, expensively, and with a large dose of racism.
So we’re talking about trading off your concerns over potential abuse and misuse of technology vs concerns about potential abuse and misuse of police powers vs just allowing traffic to kill more cyclists and pedestrians.
The nihilistic libertarian says ‘precisely - government doesn’t work - we can’t hope to make things better by concentrating power’
An optimistic democratic-institutionalist says ‘we can constrain governments at the ballot box and with constitutional limitations’.
I suspect our disagreement lies somewhere in the chasm between these world views.
I think the point is these aren't just license plate readers. For example, are they recording fingerprintable bluetooth or other wifi signals of occupants of cars that are going by?
(To be clear: the government requires you to display a license plate on your car. They do not require that you broadcast an identifiable Bluetooth signature as you drive around. If you happen to do so - something you have a choice to avoid - it seems disingenuous of you to say ‘I don’t want anyone to take any notice of the Bluetooth and WiFi signatures I choose to broadcast’.)
i see it more like an argument against increasingly networked solutions. red-light cameras used to just upload when an incident occurred (or maybe required routine physical downloading?). the object itself did one thing. now it's cheap to feed continuous live feeds to a centralized processor... and once it's there, it's cheap to do a lot more.
Okay so is the argument that if a city wants to install red light cameras they must install analog film based systems because if they install a CCD camera with a raspberry pi and a network connection they could use it for other purposes?
Obviously the electronic networked solution is going to be cheaper and easier to operate. It would be insanely wasteful to force government organizations to use expensive cumbersome processes just to ensure they can’t be misused.
The solution is legislation and enforcement of rules that constrain what local governments are allowed to actually do with equipment they install and the data it collects, not regulation of what the equipment they install could theoretically be converted to accomplish.
the argument is that they should be called traffic surveillance cameras.
for the record: whatever they're called, and in fact even if they just POSTed txt license plate numbers, i'd still be opposed because they're collecting info about people who've done nothing wrong (unlike red light cameras).
> No one expected doorbell cameras to join a mass surveillance network
No one? There's a reason I never bought one of these things. This was a pretty obvious outcome to anyone who even slightly distrusts corporations or the government.
"Autonomous dragnet surveillance cameras" wasn't as popular with the focus groups.
> No one expected doorbell cameras to join a mass surveillance network
Doorbell cameras were at least for individual security, initially. There's no attempt being made to market Flock as anything but mass surveillance. Surely we all expect them or their successors to be detecting peoples' faces, walking gates, and who-knows-what-else in the near future.
I always define them as "Remotely managed, field-upgradeable, multi-interconnected computing devices with full spectrum cameras and other sensory capabilities. Information is monitored in real-time by increasingly capable AI technology. They have edge AI capability for intelligent classification. They can be upgraded without announcement at any time. They can be replaced in the field with higher capabilities at any time. The data they captured can be accessed at any time by anyone with sufficient permissions granted by an entity capable of granting such permissions. Not all users have the same permissions. Multiple user interfaces with differing capabilities exist. There exists a capability to aggregate data streams from multiple sources provided the user has sufficient permissions. Deleted data remains accessible to users with higher clearance."
That's cute and technically accurate but doesn't help when having conversations with the public.
"Warrantless mass surveillance technology" creates a category into which these and similar devices can be aggregated together where people's concerns tend to be about the same.
That phrase was suggested by a local chief of police while we were having discussions on policy.
"They detect Bluetooth or Wi-Fi signals emitted by mobile devices inside passing vehicles and capture the device’s MAC address, a unique identifier that allows the same device to be recognized at multiple locations"
It's simply because the language hasn't caught up with the technology yet. A similar problem happens in courts where legal terminology hasn't caught up with technology.
There's no malintent with it, it just means that the public conversation hasn't gotten to a point where these distinctions can easily be made to a wider audience. We'll get there.
The claim that there’s no mal intent to copaganda comes off as weirdly apologist.
Do you really think people don’t know how they’re framing the situation when they choose to use language this way? It’s all just an innocent misunderstanding, that just so happens to serve the interest of existing power structures?
I think you're misinterpreting my words. Of course there's malintent with copaganda.
I'm talking about the random schmos out there who don't know what "ALPR" even is. These folk are more and more learning about what these things are and pushing back on them.
We are only ever one computer science paper or github repo away from all of their moat fading to dust anyways. The question everyone financial should be asking is “what happens when folks don’t _need_ to pay these people anymore?”
Anthropic? absolutely. But there is a strong Jevons paradox component.
OpenAI, Google, or SpaceX have a moat of hardware and energy. If AI is 100x cheaper orgs will use 100x more AI and the benefit of algorithmic improvements will flow to whoever has the hardware to run it.
Each token prediction is one big function call. Then you just recursively generate more tokens until run out of context or the model predicts a next token indicating end of sequence. Technically the model outputs a matrix where the last row is a probability distribution, but I’m counting sampling from it as part of the chain. Hundreds of billions of dollars has gone into just making the function fatter and gradually changing pieces here and there.
I remember the concept of layers, as essentially defining the matrix math dimensions. And for a given model/framework, they were static. That always bugged me (not very dynamic).. is this still the case?
The sizes for the matrices are decided up front (“hyperparameters”, more or less guesstimated) and the values that are learned (“parameters”, for completeness) and remain static after training. The layering just refers to the core “transformer block” being chained inside the model that contains most of the weights. Those themselves are pretty simple, usually just an MLP followed by an attention function, with variations on formulation & “multi-heading” in most of the new models. Behind the terms and math and inner architectural choices, it’s pretty much still the same GPT pattern of ‘“lie” = decode(sample(lm_head(T_1(T_2(…T_n(encode(“the cake is a”))))))’, with T_i(…) being the ith layer.
You know, I haven’t kept up with MoE and etc where there’s a bit of selection going on, so I should probably be more humble. I think new work has only added different “paths down the same hill” though (no recurrence, just select different matrices), but could be wrong there. I don’t think I’m wrong on my general intuition, just want to be epistemically honest!
They log all DNS requests made to their public resolver in a searchable internal database, at least when I worked there a decade or so ago. I wonder if they seed their crawler with it?
DNS servers never see subpaths you request, only the domain itself, so that wouldn’t help with a hidden path. But there are lots of other ways to get it: caches/CDNs can leak paths, Chrome presumably sends Google a bunch of request details, and so on.
It’s a different story if it’s a subdomain though, OP wasn’t clear.
reply