Your devices recording you is something that doesn’t happen.
That requires a lot of bandwidth, or a lot of battery and a little bandwidth. There’s no evidence of that happening.
Honestly, the creepier thing is that they don’t have to to get creepily accurate ads.
Geolocation data means they know who you spend time with. They also know their search history. They know your interests. They can look at what people who seem similar to you search for.
Plus, they serve you a lot of ads so they can afford to have a lot of misses.
Your devices recording you is something that doesn’t happen.
There’s no evidence of that happening.
What’s your basis for this claim? I mean that might be the case, but is there some reason you’re able to have so much certainty?
I understand the “bandwidth” argument, and the “they don’t even need that” argument (I make a similar case in a comment to another reply), but neither of those support the idea that it can’t be happening.
Getting around the bandwidth problem isn’t that crazy: Low bitrate encodings (cause the audio doesn’t need to be human-comprehensible) and edge compute (i.e. doing some work on the device before sending anything) could mitigate this significantly, so it’s hardly impossible.
I think we mostly agree, I just wouldn’t apply that degree of certainty. But if you’re really confident that it’s definitely not happening because it’s definitely not possible, maybe you know something I don’t?
I guarantee you could do it all on-device with only a few KB upload per day.
Use a poor quality but fast text to speech algo, and use a bloom filter to track ad keyword hits and upload it periodically. It’ll use very little battery or data. If you’re wrong, the users don’t click and your online ad server explores other keywords.
Personal story time. A few years back, I texted (through Whatsapp) a colleague for a few minutes about a friend taking up welding as a supplementary source of income, and immediately (within the day) received a targeted ad in the Duolingo app for… welding torches. Important facts, I don’t weld, I’ve never done any welding and I don’t know anything about welding. How this bit of info got from whatsapp to whomever was providing interstitial ads in the duolingo app, I have no idea. My best guess is still that google’s keyboard app is logging every single keystroke I type and aggregating it in a database somewhere. I can’t fathom how that shit isn’t extremely illegal.
I don’t think so. Text to speech and speech to text engines are availiable locally without much power i guess. Also there could be error prone algorithm which records low quality audio for sake of performance and extended battery which is still enough. Also its not far off from recording and sending low quality audio to servers as intermet speeds are much faster than an audio stream
Your devices recording you is something that doesn’t happen.
That requires a lot of bandwidth, or a lot of battery and a little bandwidth. There’s no evidence of that happening.
Honestly, the creepier thing is that they don’t have to to get creepily accurate ads.
Geolocation data means they know who you spend time with. They also know their search history. They know your interests. They can look at what people who seem similar to you search for.
Plus, they serve you a lot of ads so they can afford to have a lot of misses.
What’s your basis for this claim? I mean that might be the case, but is there some reason you’re able to have so much certainty?
I understand the “bandwidth” argument, and the “they don’t even need that” argument (I make a similar case in a comment to another reply), but neither of those support the idea that it can’t be happening.
Getting around the bandwidth problem isn’t that crazy: Low bitrate encodings (cause the audio doesn’t need to be human-comprehensible) and edge compute (i.e. doing some work on the device before sending anything) could mitigate this significantly, so it’s hardly impossible.
I think we mostly agree, I just wouldn’t apply that degree of certainty. But if you’re really confident that it’s definitely not happening because it’s definitely not possible, maybe you know something I don’t?
I guarantee you could do it all on-device with only a few KB upload per day.
Use a poor quality but fast text to speech algo, and use a bloom filter to track ad keyword hits and upload it periodically. It’ll use very little battery or data. If you’re wrong, the users don’t click and your online ad server explores other keywords.
I don’t think they need to do this though.
Personal story time. A few years back, I texted (through Whatsapp) a colleague for a few minutes about a friend taking up welding as a supplementary source of income, and immediately (within the day) received a targeted ad in the Duolingo app for… welding torches. Important facts, I don’t weld, I’ve never done any welding and I don’t know anything about welding. How this bit of info got from whatsapp to whomever was providing interstitial ads in the duolingo app, I have no idea. My best guess is still that google’s keyboard app is logging every single keystroke I type and aggregating it in a database somewhere. I can’t fathom how that shit isn’t extremely illegal.
Instead of recording and sending to servers, the tracker could use some speech to text locally amd send transcription to the servers too
Right - that would require enough battery and processing power to make it obvious that it’s doing something like that.
How much battery are “ok Google” and “hey Siri” using?
I don’t think so. Text to speech and speech to text engines are availiable locally without much power i guess. Also there could be error prone algorithm which records low quality audio for sake of performance and extended battery which is still enough. Also its not far off from recording and sending low quality audio to servers as intermet speeds are much faster than an audio stream
They are available locally yes, but not without much power and processing.
So much power and processing usage, you would notice.