Founder Vikhyat defended the task as essential engineering work.
@GergelyOrosz don't agree tbh. data labeling sounds low status but it's actually incredibly valuable work and no one is above it
Just learned:
Software engineers used to do manual data labeling at Scale AI while Alex Wang was CEO. After he left, new leadership joined, and were HORRIFIED to learn this. Stopped it ASAP
Now at Meta, software engineers are assigned manual data labeling... see the pattern?
Positive users defend Meta assigning engineers to data labeling as valuable for model quality and understanding pipelines, while negative users call it a wasteful misuse of talent likely to drive attrition.
@GergelyOrosz manual data labeling full time, or some small % of time? having engineers do a bit of labeling to develop intuition for the task, check the labeling template, and/or to validate the quality of labels from contractors is IMO valuable
Just learned:
Software engineers used to do manual data labeling at Scale AI while Alex Wang was CEO. After he left, new leadership joined, and were HORRIFIED to learn this. Stopped it ASAP
Now at Meta, software engineers are assigned manual data labeling... see the pattern?
looking at the data is underrated; you can have a ton of impact by looking at the data and doing annotations, e.g.
Just learned:
Software engineers used to do manual data labeling at Scale AI while Alex Wang was CEO. After he left, new leadership joined, and were HORRIFIED to learn this. Stopped it ASAP
Now at Meta, software engineers are assigned manual data labeling... see the pattern?
@GergelyOrosz !! that's so wild it's hard to believe. how can that payroll math on cost of labeling possibly pencil out for them?
@__apf__ Fulltime! Forcefully assigned. It's why so many devs at Meta are actively searching for new jobs. There's around 5,000 of them reassigned for FT data labeling
I don't think the lesson here is "engineers shouldn't be labeling data"
Just learned:
Software engineers used to do manual data labeling at Scale AI while Alex Wang was CEO. After he left, new leadership joined, and were HORRIFIED to learn this. Stopped it ASAP
Now at Meta, software engineers are assigned manual data labeling... see the pattern?
@vikhyatk @GergelyOrosz i dont think this is a weird thing to do particularly for a data labelling startup. prolly the single biggest failure mode for a data startup is that the SWEs get detached from the data process
( i work in a data startup and i like data labelling)
@GergelyOrosz don't agree tbh. data labeling sounds low status but it's actually incredibly valuable work and no one is above it
casual saturday
@GergelyOrosz don't agree tbh. data labeling sounds low status but it's actually incredibly valuable work and no one is above it
@peekknuf At Scale AI, actual sw engineers hired to build stuff then got to do mostly data labelling. Now they are building stuff at Scale
1. Bad for fulltime software engineers - they HATW doing it
2. At Scale AI, new leadership couldn’t believe devs were so under-utilised for something that was marginal gain at best and so pulled the plug on this weird practice
3. You get just as good if not better results w contact labelling with select folks as I hear
@ShcChy @GergelyOrosz when you outsource to the lowest bidder you get shit data, which leads to shit models. the frontier labs realize this, which is why they're spending billions of dollars to acquire higher quality data
@damoosmann AI reply, and this account only has AI replies
Blocked
@__apf__ Fulltime! Forcefully assigned. It's why so many devs at Meta are actively searching for new jobs. There's around 5,000 of them reassigned for FT data labeling
@GergelyOrosz > manual data labeling
Why is it bad? I don't understand the context: if you want to train the model, why you cannot manually label the data (at least partially)?
@ArturTanona @GergelyOrosz Where to start? How about the fact that many highly skilled Meta engineers have been told to data label, which is a task normally given to less skilled contractors.
The data labelling is part of the grander plan to replace these same engineers with AI, using this labelled data.
@GergelyOrosz Scale AI整个公司就是靠人肉标数据撑起来的。
肯尼亚那帮标注工一小时挣一两美金，给OpenAI标最脏的内容标到做噩梦，这事Time都报过。
王走到哪都让人手动标数据，不是什么怪癖，是他从第一天就吃这碗饭。
新领导惊恐，只能说明他们没搞懂这行最值钱的就是人盯出来的那点干净数据。
@GergelyOrosz Were these actual software engineers who used to ship real code or "swe" who were actually hired to do the (mildly technical) labeling to begin with so that they later could pad a CV? not saying this was the case but data labeling business is just like that...
@GergelyOrosz Getting users is the part I skipped. Never paid for a single one.
The app spread because my family kept showing it to each other.
That only works if people keep opening it on their own. You can't buy that.
@GergelyOrosz This is objectively not true, margins would be awful
@GergelyOrosz depending on the data, if code related, for a correct labeling I wouldn’t assign anyone else than software engineers. Who else is more suited then?
@vikhyatk @GergelyOrosz yeah like you actually want experts in a thing to perform data labeling in that thing
@GergelyOrosz Scale's new leadership overcorrected. Domain labeling by domain experts is load bearing, not embarrassing. Meta relearned this the hard way.