• Fondots@lemmy.world
    link
    fedilink
    arrow-up
    24
    arrow-down
    1
    ·
    3 days ago

    I work in a 911 dispatch center, one time I had someone from the police in one town, let’s call it Townsville, trying to get ahold of an officer from another town in our county, Citysburg.

    I go to put a phone call request in for them, and ask what it’s regarding, and they say it’s about an incident that happened at the Mega Lo Mart in Citysburg.

    There is no Mega Lo Mart in Citysburg, but there are a couple in nearby towns, some of them have Citysburg mailing addresses, or people might casually say that they’re in Citysburg because they don’t really know where the borders are, this is pretty common and we deal with it a lot, so I ask if they have the address to make sure that I’m getting them in contact with the police department that actually covers that store.

    They spit out an address like 123 Main St in East Jabip

    East Jabip isn’t in our county, I’d never even heard of that town before. I punch it into Google maps, it’s like 2 or 3 hours away from us and sure enough there is a Walmart at that address there.

    So to make sure I wasn’t missing something, I asked why they wanted to speak with Citysburg police if the incident happened in East Jabip.

    And they reply “yeah, I’d never heard of east Jabip either, so I punched the address into Google and the AI told me that it was in Citysburg”

    Just blatantly false, AI-hallucinated bullshit.

    And our cops (or their office staff, can’t remember who exactly it came from) just blindly believed it and didn’t bother to verify it at all.

    And Citysburg isn’t some nowhere town, we’re a fairly dense suburban county, these two towns are maybe about a 20 minute drive from each other if traffic cooperates, and Citysburg is our county seat, cops from all over the county are there all the time for court and such and the surrounding areas, they should know that area at least well enough to know that East Jabip isn’t there

    And they even admitted that it didn’t sound right to them, but they still went ahead with it and didn’t question the AI.

    And luckily this wasn’t for anything too urgent, it was for credit card fraud or something along those lines. But actual emergencies get called into us all kinds of bass-ackwards ways like someone in another state calls their mom who lives here who calls her local police who transfer her to us so that we can transfer her to the police where the person who’s actually having an emergency is.

    That kind of stuff happens pretty frequently by the way, that’s not some bullshit scenario I’m making up, I get that probably on a weekly basis.

    And so it burns me up thinking that if there had been an actual emergency, having to sort out this bullshit could have caused delays in getting help to someone who needed it because someone outsourced their thinking to a shitty AI

    • MadeInDex 📰🌎@lemmy.worldOP
      link
      fedilink
      arrow-up
      3
      ·
      2 days ago

      And luckily this wasn’t for anything too urgent, it was for credit card fraud or something along those lines. But actual emergencies get called into us all kinds of bass-ackwards ways like someone in another state calls their mom who lives here who calls her local police who transfer her to us so that we can transfer her to the police where the person who’s actually having an emergency is.

      Thank you for giving us so much insight and taking the time to explain the background! I haven’t even considered this kind of scenario, I mean this sounds like a patter and something that might even increase as more people rely on bad AI. Can you maybe raise this issue with a higher up office? Also thank you for the important work you are doing, if I had an emergency, I would hope somebody considerate like you is on the other end! :)

      • Fondots@lemmy.world
        link
        fedilink
        arrow-up
        2
        ·
        2 days ago

        It hasn’t become a pattern yet, at least not in my area, but I’m sure it’s not gonna be the last call I handle where bad AI info is gonna be a problem somehow

        I did mention it to my supervisor, and did my best to politely and professionally chew the cop out for blindly trusting the AI search results.

        We’ve already had plenty of experience sorting out issues where addresses and such that people got from Google maps or whatever don’t match up with reality, but we’re pretty good at catching that kind of stuff and figuring it out.

        As far as getting calls passed along to us in weird ways, unfortunately I think the only solution for that is for people to just suck it up and call 911 on their own instead of calling their mom. With a few exceptions for VoIP phones and such, if you call 911, it’s going to your local dispatch center wherever you’re located, and we hopefully have some kind of approximate location for you. The amount of times I’ve had to play 20 questions with a 3rd party caller who has no clue about what’s going on for something I could have cleared up in about 30 seconds if someone had just called themselves is pretty insane.

        And even if you do have to be that 3rd party making a call for someone somewhere else, calling 911 is probably going to be your fastest way to get there. It can be weirdly hard to find a good number for local police sometimes, but most 911 centers have access to some database or service to find the right contact info faster than you probably would googling it (I’ve gotten calls come into me for towns in other states or even other countries, because they had the same or similar names to ones in our area, and the caller didn’t double check that they were calling for the right Townsville)

        Although, I will say, some 911 centers are pretty terrible. One that borders my county has a bad habit of transferring any calls for something that isn’t in their area to us, either because they didn’t bother to verify the location, or because they just can’t be bothered to look up the information themselves and they know we’ll do it for them. Occasionally they even transfer us calls that they should have kept themselves and I have to transfer a really frustrated caller back to them.

        That’s another thing my higher-ups are aware of and working on.

        • MadeInDex 📰🌎@lemmy.worldOP
          link
          fedilink
          arrow-up
          1
          ·
          4 hours ago

          Wow that’s so interesting, thank you for the insight of what’s going on in the background of such a call! It really sounds like most of all you need “patience” for this job ;) I have always been wondering: if somebody calls you and just breathes heavy or stops talking like they passed out or something. What happens next? And how accurately can you figure out where somebodies phone is for example?

          • Fondots@lemmy.world
            link
            fedilink
            arrow-up
            1
            ·
            1 hour ago

            Policies and what resources are available are going to vary a bit from one agency to another, but assuming it came in to us on a 911 line

            From landlines, we get an address for the phone number. There’s a couple exceptions to that with certain kinds of business and VoIP lines where the address we get may not actually be the actual address where the person is, or there’s always the chance that the phone company has wrong info, but generally speaking if you call from a landline we know where you are.

            From cell phones, things get a bit fuzzier. For the most part we’re relying on triangulation from cell towers to locate you (we call it “Phase II”) which means the quality of that location can vary from pretty good to basically useless based on how many towers your phone can reach, signal strength, geography, etc.

            What that location looks like is we get a set of coordinates

            An “uncertainty radius” or “confidence factor” which is a distance in meters from that point that the caller is probably within

            A “confidence percentage” which is how confident the system is in that location (I’ve literally never seen this be anything other than 90%)

            So what it ends up looking like is something like “90% confident that the caller is with 200m of 40.12345°N,-90.12345°W” (random-ish coordinates, not sure where that location actually is, but it’s definitely not where I work)

            I’ve seen the confidence factor be in the single digits, and I’ve seen it in the thousands. Sometimes it takes a minute before we get a good fix, sometimes it comes in right away, sometimes we never get a good location from it.

            My agency’s policy is that if we have a confidence factor of 300 or less, we can enter the call as normal with just that phase II if we’re unable to verify that location any further

            And if they’re in somewhere like a wide open field or parking lot or something, 300m is pretty good, they’ll probably see you when they get out there. If you’re in a denser neighborhood with apartment complexes and a bunch of houses, wooded areas, etc. that’s really not much to go on. Usually we can get at least that 300m, but again not always.

            That phase II location also takes a while to update, if we’re lucky we can only get an updated location every 20 seconds or so, so if, hypothetically, you’re in a car flying along the highway at 70mph, you could be about a half mile away from where you were by the time we got a new ping.

            So we always try to verify the location, and we can’t, as my callers like to put it “just GPS your phone”

            New technology is rolling out, we can sometimes get actual GPS locations from your phone which is usually more accurate and updates faster, but it depends on what settings you have enabled, what your carrier supports, etc. I think my center currently can only get it from iPhones. Same for your emergency information like contacts, medical info, etc if you’ve filled that out.

            Once you hang up with us, that’s usually pretty much it, we’re not getting any further updates on your location even if we call you back and you answer.

            We also don’t get any of that if you call on a 10-digit non-emergency line, usually we get your phone number and maybe a name on the caller ID, but depending on how the call got routed to us, like if you were forwarded from a station, we may not even get that much.

            If we get a call with no other usable location info, if it came from a landline we can look up the phone number to get the address.

            We can also look up the phone number to see if we had any prior calls from that number that we might be able to get an address from. We only store those records for about a year, sometimes our police departments have records that go further back they can look up, but we need something to go on to pass it along to the correct department that would have those records.

            Pretty much anything beyond that is usually something that needs to be initiated from the police. There are only very narrow circumstances where we’re able to request for a phone company to try to ping your phone, and even if we can do it, the location may not be any better. They can also try to get subscriber info from the company to get your home address (although that’s not always super useful, people move and don’t update their address, are on someone else’s plan, etc) if they get a name and date of birth they can try to look up your info from your drivers license info (again assuming it’s up to date) property records, etc.

            So if we get a call that’s just an open line with heavy breathing or something else suspicious, we’re using those tools to try to get someone out to at least the general area to try to locate the, and police are hopefully using whatever other resources they have on top of what we do to try to narrow it down if needed.

            We’re probably going to enter it as a hang-up call or a suspicious activity which just gets a police response unless we heard something that makes us specifically think fire or EMS are needed.

            If we heard yelling, gunshots, alarms going off, etc. then we might enter it as something else as appropriate to make sure we’re sending the right resources.

            If they stop talking to us while we’re on the call, hopefully the first thing we got from them was a location, it’s the first thing we ask, otherwise all the same thing applies.

            If it’s just an open line, we’ll stay on for about 30 seconds or so to see if we hear anything. If we don’t we enter it as a hang up, try to call it back, and if they don’t pick up we just kind of move on and it’s in the hands of the police to do something about it.

    • Ænima@lemmy.zip
      link
      fedilink
      arrow-up
      3
      arrow-down
      1
      ·
      2 days ago

      Boomers have some of the highest lead levels of any currently living generations. They are also, on average, very gullible. Them holding positions of power, during genAI startup, is why ICE was arresting citizens, with multiple forms of ID, because an app told them the person was someone else. They are told to believe the tech and they don’t question it because they don’t question anything someone with authority tells them.

      • Fondots@lemmy.world
        link
        fedilink
        arrow-up
        2
        ·
        2 days ago

        As much as I’m happy to blame boomers

        The youngest boomers are in their 60s now, not too many of them are out working as any kind of field agent. Most of those assholes out there now are Gen X, Millennials, or even Gen Z.

        There have been some changes under this administration, and I don’t know what the current “rules” are (as if they care about following rules anyway) but I know at one point there was actually a mandatory retirement age for federal LEOs, and I’m pretty sure it was at about 55 or 60, so under those old rules you couldn’t be an ice agent as a boomer. Some probably hung around in admin positions and such, and since the rule changes I’m sure a couple have come out of retirement, but I’m pretty confident that that’s a vast minority.

        The issues with not questioning authority, gullibility, tech-illiteracy, etc. remain though.

  • Zedstrian@sopuli.xyz
    link
    fedilink
    English
    arrow-up
    25
    ·
    3 days ago

    It’s just gluing together data from random sources it finds with no care for the accuracy or recency of that data; Lemmy has about 38,000 MAU, not 77,000.

    • Aneb@lemmy.world
      link
      fedilink
      arrow-up
      6
      arrow-down
      1
      ·
      3 days ago

      The big difference is intent. Politicians purposefully lie to further an agenda paid for by super PACs. AIs are dumb LLM text prediction machines, they lie because they are slop. That’s actually something they share again, they’re both shit.

  • LumpyPancakes@piefed.social
    link
    fedilink
    English
    arrow-up
    13
    ·
    3 days ago

    Lies a lot about Home Assistant I think, but that’s probably because the info online also doesn’t match reality. Garbage in garbage out still applies.

    • TimeSquirrel@kbin.melroy.org
      link
      fedilink
      arrow-up
      7
      ·
      3 days ago

      I also use Homeassistant. It’s a very rapidly moving target, development-wise, it’s understandable that a website listing a procedure to do something from a year ago may not be correct today.

  • stumu415@lemmy.zip
    link
    fedilink
    arrow-up
    9
    ·
    3 days ago

    Browsing is so much worse now. Note I don’t use Google but Duck and turned all the AI shit off. Before Duck also had an AI overview and so many times I had to find actual different sites to verify the information. Wasted so much time. AI truly made everything worse. I had such a simple app to control my curtains. Now it’s fucked by adding an AI POS assistant that takes up the whole main page.

    Urgh I hate this enshittification in the name of progress.

  • Riskable@programming.dev
    link
    fedilink
    English
    arrow-up
    1
    arrow-down
    15
    ·
    3 days ago

    Meh. For a control I need to see a comparison of the lies in the search results. I bet Gemini is doing a better job than the limitless bullshit that people are expected to click on.

    Let us not forget that Gemini is often pointing out basic facts like, “vaccines actually work” when people search for information about them. This angers a lot of complete idiots and results in anger-inducing clickbait articles… Like this one.

    The fact that a machine can get 90% of everything correct is fucking amazing. And the pace of improvement in AI in general is astounding. Try to keep that in perspective when reading stories like this.

    Anything over 50% is probably better than a random sampling of humans could do, based on our current state of education and “knowledge upkeep” (which is something most humans fail at, utterly and completely).

  • Kairos@lemmy.today
    link
    fedilink
    arrow-up
    2
    arrow-down
    8
    ·
    3 days ago

    Not really. Something that cant intentionally do anything can’t really lie.

      • The_Decryptor@aussie.zone
        link
        fedilink
        English
        arrow-up
        3
        ·
        3 days ago

        To me lying implies an intent to deceive, LLMs can’t do that as they have no intentions or understanding of the output they produce.

        It’s not lying, because it’s also not telling the truth either, it’s just statistically weighted noise.