Seriously I’ve seen a guy see a bunch of pixels and go “Idk but that reminds me of Mexico” and he was right. There’s no way the three letter companies wouldn’t want that kind of skillset, right?
Who says they don’t already? I knew a guy who worked for a major government agency who’s job was to look over horrible pictures to try to figure out where they were taken.
I bet the most skilled people do work for them, not just youtube. But talking to that guy, it’s a tough job, not for everyone. Instead of a random streetview image it’s a scene of abuse, so…
I just assume they’re super lucky. Or that they play so much, they recognize when it shows them the same picture because it doesn’t have an infinite set of images.
You should check out Rainbolt’s Geoguesser runs. He often narrates his thought process as he goes. He’s looking at things like the angle of the sun and shadows to determine what hemisphere of the planet he’s on, looks at the vegetation and soil types to further pinpoint his longitude, etc. He’ll even take photos sent directly to him from viewers, and he’ll find out exactly where the photo was taken, even if that image had never existed on the internet prior.
Some of these guys are insanely talented at this. Rainbolt is probably the most entertaining and educational, IMO. Definitely worth checking out some of his videos.
Like the other guy said, it’s street view. So the number of images isn’t quite infinite, but it’s high enough that just memorizing the images isn’t realistic.
Because of that, though, road patterns are a solid chunk of what the really good players are looking for/at. There’s definitely more to it than that, but because of the nature of the game there’s almost literally always a road in view.
My guess is that some memorization exists, check the geoguesser grand finals, there was one play where both players marked the exact same spot with absolutely no clues another that “maybe Mongolia”.
There’s definitely memorization, but there’s just a ton of tricks that aren’t explicitly memorizing the images themselves.
They can sometimes narrow down the area based on the car that’s taking the images, or how wide the lens is. As a made up example, if they see a red toyota corolla is the car taking the photos- which they can tell based on the mirrors appearing- they know it’s specifically mainland Malaysia.
Very small context based clues a lot of the time.
It has a practically infinite set of images. Seriously, I’ve seen people do geoguesser and get nothing but back roads in the middle of basically Anywhere, Earth; I’ve seen people get stuck inside malls, go into caves, walk down random-ass rabbit trails in the middle of a forest surrounded by miles of fucking nothing (I wish I could remember what video I saw that in, it was fucking insane). Geoguesser pulls from the full catalog of Google Street view images, and there are a lot of street view images.
Really it’s crowdsourcing and statistics. Show an image to a big enough crowd and someone will pick something up. It’s like the birthday problem but with geography.
There are but they’re part of the crowd: They’ve just been doing it so long they’re great at it.
A thousand people can take guesses, but only a skilled player can say “Oh yeah, that foliage looks like San Juan or some other mountainous region of Argentina”, or “That black electrical tape is a Nigeria thing”
People who live in those places aren’t likely to notice how their electrical poles are different.
It’s 100% not about crowdsourcing. Idk how that even makes sense- do you know what Geoguessr is?
There’s some people that are absolutely cracked at geoguessr. Rainbolt, for example.
Not to pile on here, but this is not an instance of the birthday problem.
The birthday problem would kick in if we asked “what are the chances that any two of these N people know the same place, whatever it may be.”
But instead we’re discussing “what are the chances that one of these N people recognizes a specific place P.”
Edit: maybe I’ve missed your point actually — were you saying that there are many details in one image, and the chances of some player recognizing one of those details is an instance of the birthday problem?
there are many details in one image, and the chances of some player recognizing one of those details is an instance of the birthday problem?
That would be a valid model. But you are still right that it doesn’t apply: It would give the effect that a different geoguesser would get the picture right every test, while we are seeing consistent results from the top geoguessers.
I see your point, but the Birthday Problem would apply differently.
It is the chance of “collision” between randomly picked elements from two large enough sets of comparable random data. If I understand correctly, the random data here would be “geographical fact” like bush density and road width. Set A is geoguessers’ geographical knowledge, and set B is pictures’ geographical features.
So if we picked hundreds of random picture and hundreds of geoguessers and asked them, the chance of one guessing one image is high. And the person would be largely dfferent every time.
In this case, we can give one specific geoguesser a large amount of pictures and that same geoguesser would get most of them right.
The problem is getting the pictures to work from. People who don’t want to be found aren’t giving the CIA pictures of their locations, and I’d imagine the CIA already knows what road signs look like in different countries, or where certain mountains and other features are.
The problem is getting the pictures. It’s intel collection, not processing.
the CIA already knows what road signs look like in different countries, or where certain mountains and other features are.
I mean “the CIA” isn’t a person. Someone there does this work. Which is what we’re talking about.
I’m sure that those agencies have been contacting top players for that game over the last few years. But if those simply dont want to work for them, thats moot.
