As the author alludes to, the majority of this traffic will be easily discarded. It’s the 1-2% that they can’t be totally sure about which will cause them to either keep a bit of bogus data about you, or accidentally discard a bit of real data about you
To take an analogy from image processing
Say you have a perfect, crisp image of yourself. That’s analogous to (company) having a totally pristine dataset about your activities.
Now, say, you introduce some uniform random noise to the image (this is the base level functionality of the Fauxx app without any of the other layers on).
Image denoising algorithms are pretty darn good at getting a bunch of it out, but you have to decide on how aggressively you want to denoise. Denoise too little, and some noise remains but you can be fairly sure you’re not also losing some of the original image. Denoise too much, and you can be more sure that you’ve removed all the noise, but you risk losing data in the actual image.
So, this app on its own isn’t going to take the average Joe’s public footprint and make them disappear behind a cloud of random bullshit, its effect would probably be more akin to a pane of very lightly frosted glass, once (company) is done discarding the obvious noise.
As part of your overall privacy approach I think its valuable, and once its running in the background you’ll basically not notice it’s there. But nobody should be considering this as a magic bullet privacy solution.
Re: them being able to detect and remove noise based on your existing data they already have by training on it - they totally could, and depending how much data they already have, they might have excellent results in doing so. However, performing that process is costly on resources and time (not for one user, but generally). It is likely more efficient for them to take rather dumb approaches to removing noise that are not as effective but cost less and run faster.
As the author alludes to, the majority of this traffic will be easily discarded. It’s the 1-2% that they can’t be totally sure about which will cause them to either keep a bit of bogus data about you, or accidentally discard a bit of real data about you
To take an analogy from image processing
Say you have a perfect, crisp image of yourself. That’s analogous to (company) having a totally pristine dataset about your activities.
Now, say, you introduce some uniform random noise to the image (this is the base level functionality of the Fauxx app without any of the other layers on).
Image denoising algorithms are pretty darn good at getting a bunch of it out, but you have to decide on how aggressively you want to denoise. Denoise too little, and some noise remains but you can be fairly sure you’re not also losing some of the original image. Denoise too much, and you can be more sure that you’ve removed all the noise, but you risk losing data in the actual image.
So, this app on its own isn’t going to take the average Joe’s public footprint and make them disappear behind a cloud of random bullshit, its effect would probably be more akin to a pane of very lightly frosted glass, once (company) is done discarding the obvious noise.
As part of your overall privacy approach I think its valuable, and once its running in the background you’ll basically not notice it’s there. But nobody should be considering this as a magic bullet privacy solution.
Re: them being able to detect and remove noise based on your existing data they already have by training on it - they totally could, and depending how much data they already have, they might have excellent results in doing so. However, performing that process is costly on resources and time (not for one user, but generally). It is likely more efficient for them to take rather dumb approaches to removing noise that are not as effective but cost less and run faster.