Move highlights the difficulty of finding high-quality interactive training data.
See full article...
See full article...
Meta will use employee-tracking software to help train AI agents: Report
- Thread starter JournalBot
- Start date
Just train on only Zuckerberg. They're replacing him with a bot anyway, just use only him
Upvote
119
(119
/
0)
Oh hell no! I wouldn't agree to this if it's a condition of employment. If they tried to force me I would quit.
Upvote
111
(113
/
-2)
Heh. “Personal” devices aren’t safe either. Should’ve asked them about the MDM changes for BYOD and provided devices too.
Great way to surveil every single interaction employees have with family and loved ones. Probably a ton of social training data there!
Great way to surveil every single interaction employees have with family and loved ones. Probably a ton of social training data there!
Upvote
58
(60
/
-2)
They should start with Zuck and the rest of the meta executives.
Incredibly dystopian. Of course this won’t be used for workplace surveillance at all…
Incredibly dystopian. Of course this won’t be used for workplace surveillance at all…
Upvote
76
(76
/
0)
"So, we don't seem to get the requisite amount of training data out of your system. We'd like to have a chat with you, us and Human Resources. Can you come to room 313, and bring your jacket?"
Upvote
58
(58
/
0)
I think we could stop this by everyone deleting or maybe even just boycotting using Meta, Google, X, and Microsoft products and services.
I think we could do that if the associated threat was a meteor or nuclear missile, but for some reason creating AI with the intent to create mass unemployment is too abstract for people to care.
I think we could do that if the associated threat was a meteor or nuclear missile, but for some reason creating AI with the intent to create mass unemployment is too abstract for people to care.
Upvote
55
(57
/
-2)
The median salary at Meta is ~$400,000/year. They can get away with this kind of intrusive behavior.
It starts with the insanely paid jobs and ends up with all companies in the US, unless Congress does something about it.
Upvote
6
(20
/
-14)
What's better than a petty dictator using the state to track your every movement?
A thousand thousand petty dictators using their private assets to track your every movement in the hope that they can literally replace all humans with machines.
Freedom, freedom, freedom, oye!
A thousand thousand petty dictators using their private assets to track your every movement in the hope that they can literally replace all humans with machines.
Freedom, freedom, freedom, oye!
Upvote
44
(45
/
-1)
I'm emotionally torn between the grotesqueness of the proposal and the hilarious premise that tracking mouse movements, clicks, and keystrokes is a remotely sensible path toward a competent robot helper.
In more ways than one, it tells you a lot about the pitiful value that Facebook ascribes to its employees.
In more ways than one, it tells you a lot about the pitiful value that Facebook ascribes to its employees.
Upvote
93
(94
/
-1)
They say they’re using it to train AI agents to use interface devices. I can think of at least a half dozen other evil dystopian ways this could be used. What’re the chances this isn’t a smokescreen for something worse?
Upvote
24
(25
/
-1)
Here's the end game
Step 1. Meta hones this ai training system on their employees.
Step 2. Meta pays other companies to deploy it and give Meta data.
Step 3. Meta uses the data pipeline to build an advanced employee productivity scorer.
Step 4. Meta sells it to other companies to deploy it and give Meta data.
Step 5. Meta sells this data to advertising companies to display ads to worker/consumers.
???
Profit
Step 1. Meta hones this ai training system on their employees.
Step 2. Meta pays other companies to deploy it and give Meta data.
Step 3. Meta uses the data pipeline to build an advanced employee productivity scorer.
Step 4. Meta sells it to other companies to deploy it and give Meta data.
Step 5. Meta sells this data to advertising companies to display ads to worker/consumers.
???
Profit
Upvote
39
(39
/
0)
I think it's less that mouse/keyboard tracking can provide a "sensible path toward a competent robot helper" and more that it can help with the specific use case of an AI agent translating human language commands into actions on a computer OS (when combined with other LLM tools, etc)
Upvote
0
(10
/
-10)
I would imagine it’s more about employees training their replacements in the eyes of the C suite, with surveillance as a welcome value add.
Upvote
49
(49
/
0)
ubercurmudgeon
Ars Tribunus Militum
They wanted us back in the office, supposedly to improve "synergy", but it turns out they just wanted to track our mouse movement to replace us. Or so that when a rich person tells an AI assistant to book a holiday for them it can bypass the next-gen CAPTCHA on the booking web site (which probably uses cloud computing rented from the same damn corporation.)
Today, for the first time, I saw a job advert that implied a similar mandatory AI usage target. FUCK. THAT.
Today, for the first time, I saw a job advert that implied a similar mandatory AI usage target. FUCK. THAT.
Upvote
42
(43
/
-1)
I'm sure they tried, but seeing as he's already riding the hairy edge of model collapse, it probably resulted in disaster.
Upvote
14
(14
/
0)
If you're paying employees $400,000 a year and you still feel the need to watch their mouse cursors move, you badly need to revisit your hiring process and your employees' actual, real, tangible, deliverable work product.
Effective, well-run companies do not do this. Not because they are necessarily a nice and cheery big happy family, but because this is a stupid use of supervisors' time.
Upvote
55
(57
/
-2)
You left out the part where your elected leaders...are passing laws to make it illegal to stop the tech bros.
Alas, not in my lifetime, in this country.
Upvote
3
(7
/
-4)
And that's median. IC6+ can and do have seven-figure packages.
Lots of folks there are basically counting down to F.I.R.E. goals. Meta's absolutely going to lose people, but leadership's gone all-in on the bet that they can train AI replacements faster than that will happen.
This is capitalism's absolute end stage. Last chance to get out of the pit before the ladder gets pulled up forever.
Upvote
25
(28
/
-3)
Double bonus, cause they can use this extremely invasive tracking to justify laying off half of those employees for being below average in some dystopian clicking metric
Only half joking ofc..
Only half joking ofc..
Upvote
18
(18
/
0)
Who says there is significant supervisor's time required? The AI monitors and provides a report, and the supervisor then has to decide to fight the recommendation with their supervisor, and no doubt take a hit on the report about them, or give in and have the joint meeting with HR. lose-lose-lose, shareholder-value, yada-yada-yada.
Upvote
2
(4
/
-2)
What if the supervisors are getting replaced by AI too?
https://fortune.com/2026/03/14/metas-ai-team-50-flat-management-structure/
Upvote
12
(12
/
0)
Hell No!
That's all we need to have thrown at us during a 'performance appraisal".
That's all we need to have thrown at us during a 'performance appraisal".
- We don't spend enough time on the computer.
- We waste time on the computer.
- We are inefficient in our work on the computer.
- We're not doing enough to help train the AI System monitoring us on the computer.
- And in the end : "We think the AI System/Computer can do a better job than you."
Upvote
21
(21
/
0)
Yeah that makes sense. They're probably trying to catch up to Claude in 'computer use'.
Upvote
-5
(0
/
-5)
This is where all Meta employees can help our society get better simply by doing their daily work for a different, less evil company.
Upvote
21
(21
/
0)
Because this won't be used to make a computer beat the "Are you a human" tests so that now I have to do advanced calculus within a specified amount of time to order a burrito.
Upvote
13
(13
/
0)
Well stated. My first reaction was also extreme hilarity. People say "focus on the goal, not the task" whereas this is the opposite. That somehow aping all the details, the process, will give insight on the whole.
I'm sure it will work great on solving already solved problems.
Upvote
12
(12
/
0)
No, no, no! We shouldn’t be trying to make AI more human. It’s already getting hard enough for websites to tell the difference.
This reminds me of a frustrating shopping experience I had a few weeks ago. I wrote a small PHP script to turn my grocery list into a series of search links for a store's website. My goal was to save time by building the exact shopping experience I wanted. But as I clicked through the links and added items to my cart, the site flagged me. Soon, I was forced to complete a CAPTCHA for every single item!
I was punished because my behavior didn’t match their "expected" human pattern. Training generative AI to mimic humans more closely will only make this worse, eventually turning the web into a minefield of bot-detection hurdles.
This reminds me of a frustrating shopping experience I had a few weeks ago. I wrote a small PHP script to turn my grocery list into a series of search links for a store's website. My goal was to save time by building the exact shopping experience I wanted. But as I clicked through the links and added items to my cart, the site flagged me. Soon, I was forced to complete a CAPTCHA for every single item!
I was punished because my behavior didn’t match their "expected" human pattern. Training generative AI to mimic humans more closely will only make this worse, eventually turning the web into a minefield of bot-detection hurdles.
Upvote
21
(21
/
0)
I'll jump on that bandwagon. I've pretty much already done that.
Meta is explicitly blocked and has been for about a decade. I see nothing from them.
X, the same, and that started BEFORE Musk bought it.
Microsoft got kicked to the curb last year when I replaced the last MS-enabled device in the house. We literally have no connected devices made by or running anything Microsoft.
The only sticking point is Google, which I still use for searching and e-mail.
But I use the udm14 search option for Firefox that adds that "no AI" part to the searches, so search results still looks like they did 10 years ago. All ads and scripts are otherwise blocked with NoScript and uBlock. I also don't keep my e-mail ON Google, or log into my accounts (without explicitly logging out).
My e-mail is downloaded to an e-mail client in the Google account, but them moved and stored in a local folder, automatically deleting it from the Google Inbox. Every first of the month, I go in and purge any e-mails they have collected. That's just the one place now that they seem to stick everything in without necessarily parsing it by type (sent, received, spam, etc.). I spent a weekend dealing with getting about 50,000 e-mails over the last 25 years moved and saved locally before deleting them over in Google.
Also, my phone has no accounts on it, so Google can't track that. Using my phone for surfing or e-mail is, well, hard on the eyes, and I'm never in that much of a hurry to have to have it right there. If I need something from google, I temporarily add the account, then delete it again when I'm done. Yes, a few more steps, but the privacy peace of mind is nice.
Not perfect, yes, but I don't have a lot of ambition to be changing things. I figure I'm throwing wrenches in some of their gears. I'm good with that.
And yeah, the whole AI thing needs to die.
Upvote
13
(14
/
-1)
Upvote
4
(4
/
0)
Upvote
7
(7
/
0)
Ironic since zuck is the one that set billions of dollars on fire for the metaverse but need to monitor employees day to day activity?
Hilariously out of touch.
Hilariously out of touch.
Upvote
13
(13
/
0)
Not that I would ever be employed by a firm that would use this, but I wonder if my absent minded and erratic trackball movements throughout the day might actually offer poison data in this case.
Also is Meta not worried about their own proprietary and sensitive data inadvertently getting swept up in this? Even if not being directly applied as training data, just gobbling up and storing incidental passwords, SSH keys, HR docs, planning presentations etc. in a giant heap somewhere, seems insanely risky and should make a compliance manager's head explode.
Also is Meta not worried about their own proprietary and sensitive data inadvertently getting swept up in this? Even if not being directly applied as training data, just gobbling up and storing incidental passwords, SSH keys, HR docs, planning presentations etc. in a giant heap somewhere, seems insanely risky and should make a compliance manager's head explode.
Last edited:
Upvote
19
(19
/
0)
So what happens if you kind of aimlessly move around your mouse? How will that impact the training data?
Is the purpose just to provide data to the LLMs on how different buttons, drop downs, menus look and how to interact with them no matter the context?
Or is the purpose of the training data to help LLMs to perform certain actions? In that case wouldn't it require the user to somehow to provide context to the training data with for example "I am now going to a search engine and searching for <SOME PHRASE>" followed by you doing that action? What happens if you have that context but then in the middle of it you get side-tracked and click an ad? What if a lot of people have a propensity for clicking ads will that mean that the LLM will be more likely to also click on ads?
I have so many questions as to how this is supposed to be valuable and provide something useful for training LLMs.
Is the purpose just to provide data to the LLMs on how different buttons, drop downs, menus look and how to interact with them no matter the context?
Or is the purpose of the training data to help LLMs to perform certain actions? In that case wouldn't it require the user to somehow to provide context to the training data with for example "I am now going to a search engine and searching for <SOME PHRASE>" followed by you doing that action? What happens if you have that context but then in the middle of it you get side-tracked and click an ad? What if a lot of people have a propensity for clicking ads will that mean that the LLM will be more likely to also click on ads?
I have so many questions as to how this is supposed to be valuable and provide something useful for training LLMs.
Upvote
8
(8
/
0)