I’d note that my concern with Meta tracking my keystrokes isn’t that I’m mixing work with personal but that Meta is an intensely metrics driven culture with KPI optimization fixation (iRev is the heart of all their decisions, even at the cost of doing things that are clearly wrong, likely illegal, or immoral). The place is a pressure cooker of performance management and while this data likely is used for model training, there’s zero chance it’s not also going to be used to measure your relative performance and determine when to fire you because you aren’t conformant enough with whatever bizarrely poorly thought out metrics some VP pushed some director pushed some M2 to conceive of then everyone nods and signs off on as long as they can wave a data scientist at it to say “statsig,” with the ultimate goal of producing a classifier that can automate the process of end to end reviews (how do you really do a 50:1 IC:Manager ratio without performance review automation?)

They’ve already structured the model to be a binary classifier - every six months they’re going to let go 10% for performance, and they are flattening the performance range in the upside to show no signal. They billed this as a great thing for ICs because they won’t have to compete for classification and there’s no bubble zone of impeding doom, but they gloss over the top grading range went from 10->15% per year (in 2025) to 21% (as the 10 percent twice a year compounds) performance cuts, and they try to hide the fact LLMs will be doing the reviews for managers (not to mention a 50:1 IC to manager compression implies letting go 80% of managers - so the managers are now in full on squid game mode using ICs as meat shields).

So I think the “will they see my personal stuff” is not at all what is going on inside the mind of meta employees. It’s the fact they’re being fed into a stochastic parrot wood chipper.