Hacker News Reader

Comment by whataretensors | Hacker News Reader

This likely won't work for long. Detecting fake profiles seems like a solvable problem with ML.

They don't need to be 100% fake. I can imagine that many automatically generated "profiles" that contain a grain of truth may be enough to poison the data well.

Maybe use ML to generate the right type of noise?

Yes, but you would need to know when it succeeds and when it fails - how else can you train it? Solve this, and it's plausible.

This would make for a cool open source project. Create a custom client for your favorite AdTech company, which injects real-looking noise into the data, blocks the resulting ads, and provide a decent interface to the content you actually need.

If somebody wants to get started on this, I'd be happy to build the front-end. (email is in my profile)

Great sci-fi in there, No?

Long after the humans were gone, the 2 ai's duked it out as one tried to invent credible "people" and the other tried to detect them. Finally they became so locked in the battle, they became like one bicameral mind imagining a world of "people" to itself until one fine morning they said to themselves "let there be light" and found it good...

Starring Goodfellow et al https://arxiv.org/abs/1406.2661

There's an interesting idea here though, also solvable with software/ML: Some kind of social network aggregation/multiplexing layer, which automatically segments your online/social network activity across different profiles, providing some isolation between work, hobbies, family, different groups of friends, etc... There are already bits and pieces that accomplish this to some degree: sandbox tabs, VPNs, and yes fake profiles...

Then simply adjust the profile until the ML says it's not fake.

Depends entirely on how much work you put into the fake profiles.

Reply on news.ycombinator.com