Gig workers are writing essays for AI to learn from
Companies are hiring highly educated gig workers to write training content for AI models.
The shift toward more sophisticated trainers comes as tech giants scramble for new data sources.
AI could run out of data to learn from by 2026, one research institute has warned.
As artificial intelligence models run out of data to train themselves on, AI companies are increasingly turning to actual humans to write training content.
For years, companies have used gig workers to help train AI models on simple tasks like photo identification, data annotation, and labelling. But the rapidly advancing technology now requires more advanced people to train it.
Companies such as Scale AI and Surge AI are hiring part-timers with graduate degrees to write essays and creative prompts for the bots to gobble up, The New York Times reported. Scale AI, for example, posted a job last year looking for people with Master's degrees or PhDs, who are fluent in either English, Hindi, or Japanese and have professional writing experience in fields like poetry, journalism, and publishing.
Their mission? To help AI bots "become better writers," Scale AI wrote in the posting.
And an army of workers are needed to do this kind of work. Scale AI has as many as tens of thousands of contractors working on its platform at a time, per the Times.
"What really makes the A.I. useful to its users is the human layer of data, and that really needs to be done by smart humans and skilled humans and humans with a particular degree of expertise and a creative bent," Willow Primack, the vice president of data operations at Scale AI, told the New York Times. "We have been focusing on contractors, particularly within North America, as a result."
The shift toward more sophisticated gig trainers comes as tech giants scramble to find new data to train their technology on. That's because the programs learn so incredibly fast that they're already running out of available resources to learn from. The vast trove of online information — everything from scientific papers to news articles to Wikipedia pages — is drying up.
Epoch, an AI research institute, has warned that AI could run out of data by 2026.
So, companies are finding more and more creative ways to make sure their systems never stop learning. Google has considered accessing its customers' data in Google Docs, Sheets, and Slides while Meta even thought about buying publishing house Simon & Schuster to harvest its book collection, Business Insider previously reported.
Read the original article on Business Insider
