undefined | Better HN

0 pointscameldrv19h ago0 comments

The state of the art has advanced so far in doing this. I remember way back in 2017, 9 years ago now, at the Scaled ML conference, Claudia Perlich gave a presentation about using RTB data to target ads. When she got to slide 23 [1] my jaw hit the floor. This was a small ad targeting company, and again, 9 years ago. Here's what they publicly said they had:

Consumer Events:

• 100B DailyEvents

• 20+ data integrations

• Clickstream

• App usage

• Ecommerce sales

• Cash register sales

• Precise Location

Context Data:

• User

• Device

• Location

• URL

• IP

• 200 Million Devices Daily

Universal DataStore

• 50 Trillion Record Consumer History

That's about 150,000 datapoints on everyone in the U.S. For a small company. In 2017.

[1] https://cdn2.hubspot.net/hubfs/6212008/ScaledML%20Media%20Ar...

0 comments

taurath17h ago

The personalization stuff is why I avoided ML like the plague - all these folks making huge money but all of it to build a surveillance state for advertisers. Already having value my privacy enough to never work for an ad revenue company, they all seemed beyond the pale.

rdevilla17h ago

See also Dark Data from DEFCON 25: https://youtube.com/watch?v=yWqdTVQsnPg

But it doesn't need to be marketed in such a sinister fashion. In 2012 when Google Maps informed me of delays along my usual commute, complete with a GPS trace of my route home, completely unprompted, I started turning off location history (lol, yeah right). I didn't even know they were collecting that data, much less analysing it that hard.

These days, that would be considered a feature - not a dystopian hellhole, and you would be a Luddite for turning off this new smartphone augmented brain. The product will make you happy. [0]

Welcome, to City 17. You have chosen, or been chosen, to relocate to one of our finest remaining urban centers. It's safer here.

[0] https://youtube.com/watch?v=5PZ73nLZaqc

j / k navigate · click thread line to collapse