For screenwriters like me—and job seekers all over—AI gig work is the new waiting tables. In eight months, I’ve done 20 of these soul-crushing contracts for five different platforms. It’s bad.
That’s pretty accurate to my experiences. A notable point being that these projects are seemingly made by kids with zero experience in anything. It’s like investors gave 20 year olds billions of dollars and let them do whatever. Not just any 20 year old but tech bros who think they know everything.
Project requirements flip flop from day to day depending on their mood or tea leaves or the way the wind blows. One day you’re thinking you’ve followed all the project guidelines. The next day they are the complete opposite. Guess what. You’re on the hook for submitting work that no longer meets the spec.
Quality control reviewers are genuine workers or assholes or not paying attention at all. Your quality scores are what determines whether you stay on the project or get summarily removed. As the article said, you’re instantly dropped from everything without notice. You can be in the middle of working on a task and the webpage will vanish along with all your access to everything. It’s a bit of a jump scare that workers fear the most.
It’s all shrouded in much secrecy. There layers upon layers of obfuscations (read: clueless “managers” as the article said). They’re not really managers but messengers passing along the own interpretations of communications from the layer of obfuscation above them.
Still I’m 99% sure the top of the pyramid blame their shit results on us lowest peons. It becomes very obvious why LLMs are hilariously bad at times. The datasets must be so garbage given that they keep flip flopping on their requirements. Those guidelines have been passed through who knows how many layers of other workers like us. It’s a game of telephone.
The article is wrong about managers all being Gen Z. I don’t know why they put that there other than to clickbait and stoke the pointless generational war. The mid level managers are anybody from younger to middle age workers. It’s impossible to actually get contact with any of the actual engineers behind the projects.
There have been times where I’ve pointed out glaring contradictions with in project. Nobody had been paying attention at all. Not the managers. Not other workers. It’s people mindlessly reading off documents they made based on other communications they received from higher up. I was unceremoniously removed from that one project.
I was on one project where a reviewer was giving out a high rate of bad scores and entering random justifications. All of us workers spotted the trend. It took the longest time for anyone above to notice because of the layers of obfuscation.
It’s actually impossible to get in contact with anyone of consequence. The managers can only say they’ll pass along whatever concerns you might have. It the equivalent of putting your message into a suggestion box that never gets read.
I’m pretty sure most projects stack ranked. So you’re easily dropped from a project if you encounter problematic reviewers. There really isn’t any training or anything. It’s sort of a Black Mirror like world. Where workers are monkeys on keyboards sitting in individual pods. Every day a few of those pods are randomly dropped out of the system because that monkey pulled the wrong sequence of levers. Sometimes people say it’s like Severance. Nobody knows what’s really going on because we’re all kept in the dark. We just know the work is very important so we keep clicking the numbers on the screen.
Quite often the datasets are so bad, it’s impossible to annotate the data. The article pointed out the regional accent. Sometimes the images, videos, audios are so bad it’s impossible do anything with them. But it’s work that must be done so you try to invent some way to justify annotating the thing based on the guidelines. What even is the point of using that useless data but to create an LLM model for the sake of having something to show for the investors. It’s a sham.
The quality control process needs to be much better but as the article said, it’s other workers who were picked because they supposedly had better quality scores. The pay remains the same too like the article said. The whole system is a mess to begin with so they might as well be picking reviewers at random. Also keep in mind the project guidelines can change constantly. All things considered, it’s garbage in garbage out as someone already commented.
A lot of workers are grinding time to collect paychecks rather than actually care about a project. It’s a shitshow at both ends. I’ve heard workers remark on how they’ve lost faith in AI after working these jobs.
This is where a portion of the billions of investment money is going. They’re exploiting a work force to churn out datasets because they’ve already devoured the existing internet. The lowest paid I’ve seen so far is $1.50 USD for certain workers in India. On the same project, people in other countries were getting up to $20. The same projects are farmed out to different data brokers so the pay can be drastically different depending on the company.
There’s so many scammers too. Given the disparity in pay, there are people from India and Africa who will farm accounts from other regions of the world and work through VPNs. There have been speculations that some of these workers even get to be reviewers. That’s where people think is a source much of the bogus reviews that are written in broken English.
That’s pretty accurate to my experiences. A notable point being that these projects are seemingly made by kids with zero experience in anything. It’s like investors gave 20 year olds billions of dollars and let them do whatever. Not just any 20 year old but tech bros who think they know everything.
Project requirements flip flop from day to day depending on their mood or tea leaves or the way the wind blows. One day you’re thinking you’ve followed all the project guidelines. The next day they are the complete opposite. Guess what. You’re on the hook for submitting work that no longer meets the spec.
Quality control reviewers are genuine workers or assholes or not paying attention at all. Your quality scores are what determines whether you stay on the project or get summarily removed. As the article said, you’re instantly dropped from everything without notice. You can be in the middle of working on a task and the webpage will vanish along with all your access to everything. It’s a bit of a jump scare that workers fear the most.
It’s all shrouded in much secrecy. There layers upon layers of obfuscations (read: clueless “managers” as the article said). They’re not really managers but messengers passing along the own interpretations of communications from the layer of obfuscation above them.
Still I’m 99% sure the top of the pyramid blame their shit results on us lowest peons. It becomes very obvious why LLMs are hilariously bad at times. The datasets must be so garbage given that they keep flip flopping on their requirements. Those guidelines have been passed through who knows how many layers of other workers like us. It’s a game of telephone.
The article is wrong about managers all being Gen Z. I don’t know why they put that there other than to clickbait and stoke the pointless generational war. The mid level managers are anybody from younger to middle age workers. It’s impossible to actually get contact with any of the actual engineers behind the projects.
There have been times where I’ve pointed out glaring contradictions with in project. Nobody had been paying attention at all. Not the managers. Not other workers. It’s people mindlessly reading off documents they made based on other communications they received from higher up. I was unceremoniously removed from that one project.
I was on one project where a reviewer was giving out a high rate of bad scores and entering random justifications. All of us workers spotted the trend. It took the longest time for anyone above to notice because of the layers of obfuscation.
It’s actually impossible to get in contact with anyone of consequence. The managers can only say they’ll pass along whatever concerns you might have. It the equivalent of putting your message into a suggestion box that never gets read.
I’m pretty sure most projects stack ranked. So you’re easily dropped from a project if you encounter problematic reviewers. There really isn’t any training or anything. It’s sort of a Black Mirror like world. Where workers are monkeys on keyboards sitting in individual pods. Every day a few of those pods are randomly dropped out of the system because that monkey pulled the wrong sequence of levers. Sometimes people say it’s like Severance. Nobody knows what’s really going on because we’re all kept in the dark. We just know the work is very important so we keep clicking the numbers on the screen.
Quite often the datasets are so bad, it’s impossible to annotate the data. The article pointed out the regional accent. Sometimes the images, videos, audios are so bad it’s impossible do anything with them. But it’s work that must be done so you try to invent some way to justify annotating the thing based on the guidelines. What even is the point of using that useless data but to create an LLM model for the sake of having something to show for the investors. It’s a sham.
The quality control process needs to be much better but as the article said, it’s other workers who were picked because they supposedly had better quality scores. The pay remains the same too like the article said. The whole system is a mess to begin with so they might as well be picking reviewers at random. Also keep in mind the project guidelines can change constantly. All things considered, it’s garbage in garbage out as someone already commented.
A lot of workers are grinding time to collect paychecks rather than actually care about a project. It’s a shitshow at both ends. I’ve heard workers remark on how they’ve lost faith in AI after working these jobs.
This is where a portion of the billions of investment money is going. They’re exploiting a work force to churn out datasets because they’ve already devoured the existing internet. The lowest paid I’ve seen so far is $1.50 USD for certain workers in India. On the same project, people in other countries were getting up to $20. The same projects are farmed out to different data brokers so the pay can be drastically different depending on the company.
There’s so many scammers too. Given the disparity in pay, there are people from India and Africa who will farm accounts from other regions of the world and work through VPNs. There have been speculations that some of these workers even get to be reviewers. That’s where people think is a source much of the bogus reviews that are written in broken English.