Here is the original reporting https://www.404media.co/pokemon-go-players-have-unwittingly-trained-ai-to-navigate-the-world/
Good luck driving via my gps drifts as i block the phone antennas to get more kilometers. If anything, nianitic is incompetent af, i played ingress for years and i know what they are “capable of”.
Bothers me that much of this data has been collected without explaining what its intended purpose is. All those captchas that want you to pick out cars or bikes etc, Pokemon game that wants you to scan locations for free items… It simply never occurred to me that these were actually just ways to train AI surreptitiously
Never? They were very clear about the first captchas being used to train OCR, way back in like 2010.
Every day that I play Pokemon Go, I’m aware that I’m making a Faustian bargain for motivation to take a walk in exchange for my privacy.
Guess i was asleep that day 😬
It’s strange that Niantic is doing this when google has been doing the same thing with its own street map data.
Seems weird to make a competing product when (I thought) they are on good terms.
Data is always worth something, eh.
Anyone who has ever touched their software knows how this will go…
Cliff notes?
And if you’re having that much trouble with regular software, imagine how it would go with a model, which even capable programmers don’t (always) fully understand…
Make a right turn, cut through Mr Robinson’s back yard.
I’m sorry Mrs Robinson! OOOOOHHHHH!!! I am for real!!! Never meant to make your family cry, I cut through your backyard a million times!
What do we need this for? Not clear what it does
AI delivery bots maybe? It’s basically an aggregate of “here’s where it’s possible/common to walk” so it’s not useful for driving/flying AI. Also useful for marketing, knowing where foot traffic is.
The article says they’re treating it as a Large Geospatial Model (like a Large Language Model), so it seems like you could use that as a predictive way to navigate between two points. With an LLM it spits out phrases based on context. The LGM would return paths based on context.
Seems to be a tool for taking an image and Geo-locating it, even in cases where other sources of data like Googles street map cars are insufficient.
The issue is it’s going to be trained only on places people go. My understanding is that most geolocating systems are more focused on the uncommon or less traveled locations. Maybe I’m wrong, but I guess they’re probably tested it out before announcing it.
Maybe they will start/already have putting pokemon in locations that they need data for to make a more comprehensive map.
Seems questionably reliable as an independent system. Maybe could help refine more traditional systems? Not sure I understand a use case for this.
It definitely has applications in defense/Intel spaces. Knowing where a photo was taken can be really useful info.
It’s gonna be a bloodbath
