00:00When chatbots become smarter, funnier and more human-like, chances are they're being trained
00:06by human freelancers. There are at least hundreds of thousands of them around the world,
00:11often working from home, cafes, libraries and shared offices. Their job is to rate the AI's
00:17responses, write new ones and flag harmful content. What they do shapes the AI you talk to,
00:23like helping Grok tell better jokes or improving ChatGPT's math solving skills.
00:28Over the last few months, Business Insider has spoken to more than 60 AI contractors located
00:34all around the world, from Kenya to New York to Istanbul. They told us the side hustle can be
00:39flexible, rewarding and enjoyable, but also monotonous, unpredictable and mentally exhausting.
00:45Most of the people we spoke to work for Outlier, a data annotation platform run by ScaleAI,
00:50which received a $14.3 billion investment from Meta in June. The price tag shows just how
00:55lucrative the AI training industry has become. Some contractors make thousands of dollars a month,
01:00but others, particularly those in Africa and Asia, sometimes get paid just a couple of dollars an hour.
01:06Some of the tasks can take a psychological toll on freelancers, like deliberately trying to provoke
01:11the chatbots into producing violent, racist or sexually abusive responses, all in an effort to make
01:17the system safer for the hundreds of millions of people using them worldwide. While many freelancers
01:22enjoy the flexibility, it's a side hustle that comes with a lot of uncertainty. After Meta's
01:27investment in ScaleAI, many projects on Outlier were put on pause. Many contractors told us that
01:32they went weeks without any work. And the work itself is changing fast. As AI gets better at reasoning,
01:38companies are moving away from large pools of generalists, and towards smaller teams of highly
01:43paid specialists. Some platforms are paying over a hundred dollars an hour for doctors and lawyers
01:49to review prompts. There are other challenges too. Many annotators said the work feels like
01:54helping to build systems that may one day replace them. It's left many wondering when the training might
02:00be done. And when it is, what happens to the people who helped build the AI in the first place?
02:05To read the full story, head to businessinsider.com
Comments