If substantiated? Just search Onavo on HN search - I thought this was widely known for years.
If any of this is true they didn’t tell people like me about it, and at one point there were three people on the org chart between myself and the CEO.
I’m very skeptical of the allegations, but I’d be lying if I said I found them to be flat impossible. I tread very lightly on this sort of thing and I didn’t even acknowledge I’d ever heard the word Onavo until I read it on TechCrunch.
I certainly hope they’re false: FAIR seems to be the last real hope for an Open future on AI short of a complete housecleaning of the whole Valley.
But I don't think this is something he made up, it's been discussed on HN.
https://news.ycombinator.com/item?id=16381812
>I wonder if it's be possible to make a social networking startup, optimise solely for Onavo metrics, and get bought out by Facebook.
https://news.ycombinator.com/item?id=16373339
>The Onavo VPN service from Facebook is disguised as a protection mechanism but tracks the user for the benefit of Facebook.
https://news.ycombinator.com/item?id=14971839
>The database stems from Facebook’s 2013 acquisition of a Tel Aviv-based startup, Onavo, which had built an app that secures users’ privacy by routing their traffic through private servers. The app gives Facebook an unusually detailed look at what users collectively do on their phones, these people say.
I am surprised that this accusation is at all controversial.
The people in targeting/demand/supply knows absolutely nothing about profile building. And there is where all the competitive advantages lies. And also all the shady deals.
We usually keep everything very secret in profile building because that is the knowledge that allows people to leave and start competitors, but we disguise it as the usual think-of-the-children and say that profile building deals with all sort of borderline-PII and only the most vetted people should work on it.
Ask some sysadmin to list the ACL to the main ads profile HDFS or whatever it is today. it will show a couple architects who report to one SVP each.
My job was to use information retrieval, machine learning / AI, auction theory, and pragmatic statistical sampling to both accurately model and stably price ads inventory and later dollarized organic inventory to drive specific policy agendas about what got clicked on, dwelled on, commented on, seen in recommender systems in equilibrium to achieve specific policy agendas of various kinds but all ultimately tying out at top-line revenue and engagement metrics.
It did not take me long to work out that PII was useless in this pursuit, there’s no entropy in the off-property like button table as concerns CTR.
It did not take me long to realize that I didn’t want to know what it was useful for.
I easily had the seniority to run queries against Hive tables that I had an explicit personal priority of never querying.
And I left the senior leadership track at the last stop before a directorship.
I knew it sounded vaguely sketchy but you remember how many vaguely sketchy things some frisky new PM tried to get pushed through a launch card meeting only to have someone on Sheryl’s radar detonate it on the launch pad. The timeframe is the main reason I’m skeptical: Sheryl didn’t put up with crap like that she knew what was at stake.
