It is literally the “real world example” from the article.

I may have misunderstood, admittedly I just scanned it, but if you or law enforcement have to scan the universe of apps/internet to find a picture before this is useful… it’s not useful. Your starting point is a needle in a haystack.

I thought you uploaded a picture you already had, it does the scanning, and a hit might look like “some rando posted a selfie at Zilker Park 20 minutes ago on insta and that car was in the background”.

▲

kube-system 2 days ago | parent [-]

Again, the example in the article is to find the vehicle being resold online. There are only a few popular websites where people sell vehicles secondhand in any particular area, and you can easily filter to the characteristics of the car you are looking for. To search all of them is a 15 minute exercise.

Although your example may be quite viable in a repossession scenario where the possessor is known but the location is not.

▲

conductr 2 days ago | parent [-]

Right, see that is the example they went in depth on. I thought it was helping identify the chopshops and hideouts more directly as they indicated in the bullets.

This part still is the sticking point;

> When browsing Craigslist, I came across a regular car listing that showed a vehicle with buildings visible in the background. The listing claimed the vehicle was located in San Francisco. ...... Superbolt returned precise latitude and longitude coordinates that, when entered into Google Maps, revealed an exact match to the buildings visible in the listing photos.

How often do people find their stolen vehicles posted on CL/marketplace? Do police have resources to constantly browse hoping they see a similar picture of their stolen vehicles? How do they match it to the one they are looking for? Eg. if this was a cop, they may think, this vehicle matches the description of the stolen car. And this AI tells me the picture was taken at these exact coordinates (not super useful as this looks like a public place and I'm sure not where the vehicle is being stored). They still have to go out, meet the "seller", check the VIN or otherwise confirm it is the correct stolen vehicle they are looking for, then they get an arrest and recovery.

But, what if there are a dozen vehicles for sale matching said description. They now have to arrange to visit them all until they find the match or exhaust their options. How is this AI adding any value given with & without it the process looks the same; find listing, ask "seller" to meet, meet, evaluate. You don't need this AI to ask the "seller" to meet up and pretend to be an interested buyer.

FWIW, this looks like it could be a white VW Jetta to me. There are 118 in SF bay area right now just on Autotrader (granted, the hatchback is a further narrowing feature, but that's not super common either). No police department I've ever heard of has the resources to check on all these listings. If the thief stole it in SF but listed it for sale in Seattle or LA or anywhere else, how would anyone know? That's the haystack part, it's a big haystack.

▲

kube-system 2 days ago | parent [-]

Police probably do not care much unless you are in a small town. Insurance has a financial incentive to care.

> They still have to go out, meet the "seller", check the VIN

You do not have to meet the seller to check the VIN of a vehicle sitting on the street.

> But, what if there are a dozen vehicles for sale matching said description.

There might be a few vehicles in an area matching make/model/year. But it is trivial when looking at a photo to filter on further criteria... and once you look at the photo you can observe trim, exterior color, interior color, stickers, inspection sticker, etc, you will have a very high degree of certainty even on a common model.

> white VW Jetta to me. There are 118 in SF bay area right now just on Autotrader

Well yeah, because you only filtered on 2 of the dozen or so attributes that you might know.

Within a whole 500 mi of the bay area there are only 5 white VW Jetta Wagons listed. All you need to know is what year it is, to narrow it down to 2 or 3. If you know the trim, approximate mileage, any visually distinctive feature, etc, you are guaranteed a match. Even if it wasn't a wagon, it is not hard to filter down to a unique vehicle.

	▲	conductr a day ago \| parent [-]
		> You do not have to meet the seller to check the VIN of a vehicle sitting on the street. You're assuming the vehicle remains where the photo was taken. > Well yeah, because you only filtered on 2 of the dozen or so attributes that you might know. Those were the only attributes that were apparent in the photo. I said I ignored Wagon because that was a cherry picked unique filter. If it wasn't a wagon, your analysis is the same as mine, >100 vehicles in the SF bay area (I only filtered on 100 mile radius). But again, why steal a vehicle and post it for sale in the same city you stole it from? Criminals already move stolen vehicles, this is all but obvious. Basically, this helps you catch the dumbest of the dumb criminals. Someone that steals a very unique car and posts it for sale in the same area they stole it from and also leaves the car parked in the same place they took the photo. There's also a time element, if they hide the vehicle for a few weeks, then post it for sale it's more likely the initial active investigation has faded and the cops aren't actively hitting refresh on marketplace. Glad you believe this is useful, I'll continue to disagree - it might have some value but it's usefulness is being exaggerated in the article.