It will likely still be used for training, but it will spit it back censored

I think that no amount of anecdotal evidence would be enough. For a very long times doctors had anecdotal evidence that bloodletting saved patient, yet they were fooled by their bias. I’m not saying advertising isn’t spying on our microphones, I don’t know, it might be. But it doesn’t seem very plausible to me: the amount of processing necessary, and the amount of network seems way too high. Also, voice recognition is still not great currently, it was even worse years ago.

There is a difference between productivity and activity, you can be 100% active at work all day, yet 0% productive. Imagine you work on a project for 6 months and then the manager decide to drop the project. You have been unproductive for 6 months, doesn’t mean you were slacking off, but in the end when we calculate the productivity of developers, it is lower because of this.