this is extremely scary if true. are these algorithms obtainable by every day people? do they work only in heavily photographed areas or do they infer based on things like climate, foliage, etc? I would love some documentation on these tools if anyone has any.
Basically a combination of what the game geoguesser does, and public geotagged images to be able to get a decent shot at approximate location for previously unseen areas.
It’s more ominous when automated, but with only a little practice it’s easy enough for a human to get significantly better.
EDIT: yup, looks like this is the guy from the Twitter: https://andrewgao.dev/ and he’s Stanford affiliated with the same department that made the above paper and system.
I am not sure it’s the same software, but it’s a fairly good guess I think. Same software capabilities and same lab, with the same area of research.
Geoguesser is a subset of the skills used for general image geo location for open source intelligence.
In the specific cases of only using the data present in the image and relying on geographic information, it certainly does better.
Humans still do better, and can reach decent skill with minimal training, at placing images that require spatial reasoning or referencing multiple data sources.
AI tools will likely be able to learn those extra skills, but it doesn’t change that it’s the photo that’s the data leak, and not the tool. The tool just makes it vastly more accessible, and part of the task easier for curious human.
There are tons of machine learning algorithm libraries easily usable by any relatively amateur programmer. Aside from that all they would need is access to a sufficient quantity of geographically tagged photographs to train one with. You could probably scrape a decent corpus from google street view.
The obtainability of any given AI application is directly proportional to the availability of data sets that model the problem. The algorithms are all packed up into user friendly programs and apis that are mostly freely available.
It might be easier to train the AI to the specific things Geoguessr players have collected as signs that give away a location instead of letting the AI figure all those out again.
this is extremely scary if true. are these algorithms obtainable by every day people? do they work only in heavily photographed areas or do they infer based on things like climate, foliage, etc? I would love some documentation on these tools if anyone has any.
If I’m the dev, I would scrape off Google Street View with cords as data source.
deleted by creator
https://github.com/LukasHaas/PIGEON
https://arxiv.org/abs/2307.05845
Basically a combination of what the game geoguesser does, and public geotagged images to be able to get a decent shot at approximate location for previously unseen areas.
It’s more ominous when automated, but with only a little practice it’s easy enough for a human to get significantly better.
EDIT: yup, looks like this is the guy from the Twitter: https://andrewgao.dev/ and he’s Stanford affiliated with the same department that made the above paper and system.
Are you sure? The paper you linked mentioned the model beating a top geoguesser player six times in a row.
I am not sure it’s the same software, but it’s a fairly good guess I think. Same software capabilities and same lab, with the same area of research.
Geoguesser is a subset of the skills used for general image geo location for open source intelligence.
In the specific cases of only using the data present in the image and relying on geographic information, it certainly does better.
Humans still do better, and can reach decent skill with minimal training, at placing images that require spatial reasoning or referencing multiple data sources.
AI tools will likely be able to learn those extra skills, but it doesn’t change that it’s the photo that’s the data leak, and not the tool. The tool just makes it vastly more accessible, and part of the task easier for curious human.
There are tons of machine learning algorithm libraries easily usable by any relatively amateur programmer. Aside from that all they would need is access to a sufficient quantity of geographically tagged photographs to train one with. You could probably scrape a decent corpus from google street view.
The obtainability of any given AI application is directly proportional to the availability of data sets that model the problem. The algorithms are all packed up into user friendly programs and apis that are mostly freely available.
It might be easier to train the AI to the specific things Geoguessr players have collected as signs that give away a location instead of letting the AI figure all those out again.
https://arxiv.org/html/2307.05845v4
I believe this is the paper
Rainbolt has a couple of videos playing against AI. I don’t remember what they said it was trained on but it’s possible it was based on that.
ooh baby I love a good supervised learning