Meta to start capturing employee mouse movements, keystrokes for AI training • Hacker News Board

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#31

Post by pigeons » Tue, Apr 21, 2026, 6:34 PM UTC

asdff wrote: post #28

ndegruchy wrote: post #22

Earlier quoted context omitted.

The same company is trying to build an AI Zuckerberg...

It is going to be funny in a few decades when zuck transfers his shares and voting rights and estate to the ai bot, and makes himself functionally immortal. Or at least a sort of commissioned renaissance painting version of himself, probably. Imagine in 300 years we are still ruled by zuck, ellison, bezos, musk, thiel, et al, just in ai model form empowered by estates worth more than entire nations and legal protecti…

Not funny ha-ha though.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#32

Post by busymom0 » Tue, Apr 21, 2026, 6:39 PM UTC

onlypassingthru wrote: post #21

Grappelli wrote: post #18

[flagged]

Is US mouse movement different from European? Is it called sparkling mouse movement if it comes from California?

Genuine question: would right to left language based interfaces have different type of movements and thus training data than left to right language ones?

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#33

Post by anonym00se1 » Tue, Apr 21, 2026, 6:40 PM UTC

wrs wrote: post #19

>data collected would not be used for performance assessments or any other purpose besides model training And you expect Meta employees , of all people, to believe this?

In the midst of their 4th straight year of layoffs with another looming 20% cut coming, I'm guessing Meta employees are a tiny but suspicious.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#34

Post by loeg » Tue, Apr 21, 2026, 6:46 PM UTC

For context, when the article says "a list of work-related apps and websites," this includes Google properties like gmail, docs, etc, and social media websites like Facebook and Instagram, with no provision for excluding personal accounts.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#35

Post by fidotron » Tue, Apr 21, 2026, 6:49 PM UTC

Meta going all in on their brand with this.

Someone had to do it, distasteful though it may be. Could be quite hilarious what it learns in the process.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#36

Post by tmp10423288442 » Tue, Apr 21, 2026, 6:50 PM UTC

loeg wrote: post #34

For context, when the article says "a list of work-related apps and websites," this includes Google properties like gmail, docs, etc, and social media websites like Facebook and Instagram, with no provision for excluding personal accounts.

No one intelligent should be logging into their personal accounts on their work devices in any case - it's always been the case (at least in the US) that companies can do whatever invasive scanning they want on devices they own.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#37

Post by jtemplestein » Tue, Apr 21, 2026, 6:50 PM UTC

I wonder if this screen + mouse + keyboard (+ camera + speaker + mic) interface is really the right level of abstraction to model a “digital entity”

Sure, you can do everything a human can, but it also seems VERY inefficient

As an alternative, maybe you could just do network in/out?

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#38

Post by shimman » Tue, Apr 21, 2026, 6:56 PM UTC

bradlys wrote: post #24

Data collection isn’t new. The training is.

You don't think collecting this type of intimate information about your employees as a major violation of the social contract?

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#39

Post by evanjrowley » Tue, Apr 21, 2026, 6:56 PM UTC

jtemplestein wrote: post #37

I wonder if this screen + mouse + keyboard (+ camera + speaker + mic) interface is really the right level of abstraction to model a “digital entity” Sure, you can do everything a human can, but it also seems VERY inefficient As an alternative, maybe you could just do network in/out?

It's the same approach as Windows Recall, but all data remains sovereign to the company generating it.

Re: Meta to start capturing employee mouse movements, keystrokes for AI training

#40

Post by bwestergard » Tue, Apr 21, 2026, 6:57 PM UTC

arjvik wrote: post #29

jmull wrote: post #26

I like to imagine they’ll mostly capture meta employees using AIs to do work. Then they’ll deploy models trained on this, and begin capturing employees using AIs that are good at using AIs to do work. Repeat a few times and they’ll start capturing the keystrokes from people mashing their heads into keyboards with dispair and exclaiming, “Why can’t these models do anything anymore!!”

While it would be a hilarious failure mode to encounter, this is actually a good thing! These models already have the skills that humans were using them for, so either by training the models to use subagents or simply inlining the work done by the AI, you have a much easier time training the model to perform tasks from a human-distribution. The humans have done the work of making the human-distribution look more like…

Doesn't this assume that what humans are current doing with LLM agents is working out? Isn't it a bit early to bet on that to this degree?