"High-paid" is an exaggeration for many of these, but certainly a small subset of people will make decent money on it.
At one provider I was as an exception paid 6x their going rate because they struggled to get people skilled enough at the high-end to accept their regular rate, mostly to audit and review work done by others. I have no illusion I was the only one paid above their stated range. I got paid well, but even at 6x their regular rate I only got paid well because they estimated the number of tasks per hour and I was able to exceed that estimate by a considerable margin - if their estimate had matched my actual speed I'd have just barely gotten to the low end of my regular rate.
But it's clear there's a pyramid of work, and a sustained effort to create processes to allow the bulk of the work to be done by low-cost labellers, and then push smaller and smaller subsets of the data up more expensive to experts, as well as creating tooling to cut down the amount of time experts spend by e.g. starting with synthetic data (including model-generated reviews of model-generated responses).
I don't think I was at the top of that pyramid - the provider I did work for didn't handle many prompts that required deep specialist knowledge (though I did get to exercise my long-dormant maths and physics knowledge that doesn't say too much). I think most of what we addressed would at most need people with MSc level skills in STEM subjects. And so I'm sure there are a few more layers on the pyramid handling PhD-level complexity data. But from what I'm seeing from hiring managers contacting me, I get the impression the pay scale for them isn't that much higher (with the obvious caveat given what I mentioned above that there almost certainly are people getting paid high multiples on the stated scale)
Some of these pipelines of work are highly complex, often including multiple stages of reviews, sometimes with multiple "competing" annotators in parallel feeding into selection and review stages.
No comments yet.
