The points and warnings in this article would generalise to the wider population, I'm not sure if the strict AI safety researcher angle is necessary.
It also sounds like you acknowledge that even with existing, cheaper to capture and process signals (scrolling, dwelling, engagement, accelerometer, etc) it's possible to build a very detailed profile, so even without speculating about covert webcam and microphone access, the threat is there and is more immediately important to acknowledge and take steps to mitigate by forgoing centralised, large scale social media as much as possible with immediate effect.
I'm not quite sure if centralised social media sites being boycotted by AI safety researchers or even the entire Lesswrong readership will decimate Big Tech.
But even if you somehow managed to convice billions of people to decide to join the boycott and actually stick to it, social networks could rein back algorithmic curation of feeds and get people back on after reassuringly demonstrating a simple time based feed of posts by people you follow, maybe with optional filters.
Also, even if centralised social media doesn't implement any sensor tracking and anal... (read more)