5 ms·

>We filtered out all spatiotemporal edges in the entity graph with an implied speed of <300 kilometers per hour or <200 kilometers distance, IIRC. This was the

• magicalist 1y ago
  > Sounds like the bigger issue is that you're able to get "spatiotemporal" data in the first place? Yeah, this just sounds like it's written from the perspective of a data broker. Tying particular ad analytics (presumably ip geolocation?) to thousands of particular individuals and having it well populated enough to track them is "privileged first-party data access" by another name.
  • const_cast 1y ago
    Your location is leaked in many, many ways. Even if you have location services off on your phone, the first-party (Google, Apple) has access to your precise location. On Android, this bypasses VPNs, and I believe on iOS/Mac first-party apps also bypass VPNs. You are trusting that this data is not leaked to any third-parties. You cannot verify this, as the data is exfiltrated to servers which you can't verify. Okay, fine, I'll just install another operating system then, like KDE plasma mobile or GrapheneOS. Your location is still leaked 24/7. This is because your cellular modem has it's own operating system, running underneath your phone's operating system, which is triangulating your location at all times. Once again, you are trusting that telecommunications companies aren't misusing this - but please remember they're complied, by law, to make a lot of this information available to numerous third parties. Okay fine, let me just remove the Sim then and use my phone on Wifi only, always through a VPN. Your location is still being leaked potentially, for example, by your car. Your car also has a cellular modem which leaks your location, and you probably signed a contract allowing that data to be given to hundreds of third-parties. Of course, all of this is assuming you don't use any social media. Social media can also leak your location, even without location services. If you review a restaurant - that's your location. Where are your friends? You're probably around them. And on and on.
    • magicalist 1y ago
      I mean you're saying a lot for rhetorical effect, but it doesn't get around the fact that there aren't that many avenues to reliably collect this data, with high enough resolution and tied to identity, for thousands/millions of individuals, and if you do have that data, you're basically a data broker. I mean, yes, all those things are true, and they're pooled together and available for sale by data brokers. It's also disappointing that the root comment is distracting from the 4th amendment violations by making the conversation about their vague claims of selling mini-palantir demos through abusing web ads.
      • jandrewrogers 1y ago
        The assumption that the data must be "high resolution" is erroneous. Low resolution noisy data works just fine, you just need a lot more of it. You can use standard signal processing tricks such as stacking noisy low-resolution data to extract high-resolution features. This requires a lot more processing but that isn't much of a limitation. These reconstruction techniques work even if the data is from unrelated sources that aren't even trying to measure the thing you are measuring. Any data exhaust will work, people have created interesting PoCs leveraging things like HVAC data, RF attenuation, etc. High-precision weather models essentially work the same way, making inferences by stitching together diverse event data that has nothing to do with weather. High-quality high-resolution data sources largely don't exist in the way people imagine they do, so you need to do this anyway. If you have a high-resolution spatiotemporal graph for entities, tying it to identity is always trivial. It would be more common if it weren't for the fact that open source platforms scale poorly for this type of analytical processing.
    • gruez 1y ago
      >On Android, this bypasses VPNs source? >You are trusting that this data is not leaked to any third-parties. You cannot verify this, as the data is exfiltrated to servers which you can't verify. At least on Android you can theoretically disable "google location accuracy" which stops it sending nearby hotspot mac addresses to Google. That's the only public route where google gets your location without you knowingly sending to it. You also imply that mobile operating systems are surreptitiously sending locations back to google/apple even if users have all location related features disabled, but I'm not aware of any evidence this is the case, and this falls into same category as "facebook is secretly listening to you" territory until proven otherwise.
  • jandrewrogers 1y ago
    Anyone could have acquired this data the time, it was all either free or cheap. Like I said, my business was specialized data infrastructure (e.g. storage engines and analytical parallel processing), we just used these data sources for testing and demos because "free or cheap". I also have a lot of experience with privileged first-party data but that is governed by a different set of rules and is often regulated. You have to be much more circumspect about how you use it. Even though it might be convenient to e.g. slurp telemetry off a mobile carrier's backbone, what you eventually realize is the inability to do this isn't a real limitation and in some ways is a blessing in disguise.
• AnthonyMouse 1y ago
  > Otherwise it's like saying "we can figure out all stores you've been to, if we have your credit card transaction history". The preposterous thing is that payment processors aren't just allowed to collect this information and tie it to your name, they're required to do that. People talk a big game about fighting fascism, but how can you allow these laws to exist if you can contemplate what happens when actual fascists get hold of that data going back decades? They need to be dismantled now.
  • charcircuit 1y ago
    Arguing that we shouldn't do something because it would make it harder to enforce laws is not a convincing argument to me. It sounds like you want to enable people to be criminals.
    • magicalist 1y ago
      > Arguing that we shouldn't do something because it would make it harder to enforce laws If you want to do it, get a warrant.
    • vdqtp3 1y ago
      At this point in human history, is it relevant to the individual whether someone is a criminal? What matters is whether they've injured someone else. To use the US as an example (I doubt other countries are much better) it's estimated that every adult in the US commits multiple Federal felonies per day[1], Federal law is replete with ridiculous laws[2] and the number of federal laws is uncountable by Congressional Research Service staff. Does it matter at that point? [1] Three Felonies A Day - ISBN 978-1594035227 [2] https://x.com/CrimeADay https://x.com/CrimeADay
      • gruez 1y ago
        >[1] Three Felonies A Day - ISBN 978-1594035227 That's not a serious estimate: https://news.ycombinator.com/item?id=43744267 https://news.ycombinator.com/item?id=43744267
• gmueckl 1y ago
  Twitter has had timestamped amd geotagged posts for ages. Just clustering things like hashtags of tweets spatiotemporally results in a treasure trove if information about events. I'm sure that other platforms attach the same kind of info to posts. It's just a matter of scraping it.
• jandrewrogers 1y ago
  > Sounds like the bigger issue is that you're able to get "spatiotemporal" data in the first place? Almost all data is spatiotemporal data, people just aren't used to thinking about it like that. Everything that "happens" is an event with associated times and places. Tagging of events with spatiotemporal attributes, or with metadata that can be used to infer spatiotemporal attributes, is pervasive. Every system data passes through, even if not the creator of it, observes the event of the data passing through it. Event observation is not trying to track things but it implicitly and necessarily creates the data that makes tracking and spatiotemporal inference possible. These kinds of analyses rely almost entirely on knowing the events occurred; you could encrypt the contents of the data and it wouldn't matter. Software leaks spatiotemporal event context everywhere across myriad systems, internal and external, that incidentally collect it. There isn't anything nefarious about most of it and much of it is required for reasons of criminal and civil liability. What people underestimate is that you can analytically stitch together many unrelated sparse data sources with spatiotemporal attributes, many of which are quite crap or seemingly unfit for purpose, to reconstruct a dense high-quality graph. Counter-intuitively, diverse and seemingly irrelevant data sources often produce better data models. It surfaces bias, errors, manipulation, and processing artifacts in individual sources you might otherwise miss. It is much more difficult to access the obvious first-party data sources than it used to be, mostly because people with that data are far more selective about who they give access. It doesn't really matter, that is a speed bump for the unsophisticated. The exponential growth in the scale and diversity of network-connected telemetry of all types pretty much guarantees these data models will always be constructible. The historical limiter has always been the absence of data infrastructure platforms that can handle these kinds of analytics at scale.
  • gruez 1y ago
    >Tagging of events with spatiotemporal attributes, or with metadata that can be used to infer spatiotemporal attributes, is pervasive. Every system data passes through, even if not the creator of it, observes the event of the data passing through it. Event observation is not trying to track things but it implicitly and necessarily creates the data that makes tracking and spatiotemporal inference possible. >These kinds of analyses rely almost entirely on knowing the events occurred; you could encrypt the contents of the data and it wouldn't matter. Software leaks spatiotemporal event context everywhere across myriad systems, internal and external, that incidentally collect it. There isn't anything nefarious about most of it and much of it is required for reasons of criminal and civil liability. >What people underestimate is that you can analytically stitch together many unrelated sparse data sources with spatiotemporal attributes, many of which are quite crap or seemingly unfit for purpose, to reconstruct a dense high-quality graph. Counter-intuitively, diverse and seemingly irrelevant data sources often produce better data models. It surfaces bias, errors, manipulation, and processing artifacts in individual sources you might otherwise miss. That's a lot of technobabble for what essentially sounds like "there's some ad SDK that's phoning home with your gps/ip geolocation every few minutes, if you cross reference that with when flights are, you can guess what flight someone took". How far off am I? Or is there some galaxy brained AI that can infer that from disparate facts like that you stopped posting on twitter for 12 hours, your car's license plate was caught by an ALPR to be heading towards the airport, and 3 weeks ago you visited some portuguese tourism site that had an ad beacon installed?
• throawayonthe 1y ago
  but it's obviously very easy ro get from social media? e.g. you have a post from paris and then later that day a post from brussels
  • gruez 1y ago
    There's dozens of flights per day from paris to brussels, so that wouldn't uniquely identify a flight.