Now hiring: Sophisticated (part-time) chatbot tutors

There are usually two types of work for these trainers: supervised learning, where the AI learns from human-generated writing, and reinforcement learning from human feedback, where the chatbot learns from how humans rate their responses.

Update: 2024-04-13 02:15 GMT

Representative Image

YIWEN LU

After her second child was born, Chelsea Becker took an unpaid, yearlong leave from her full-time job as a flight attendant. After watching a video on TikTok, she found a side hustle: training artificial intelligence models for a website called Data Annotation Tech.

For a few hours every day, Becker, 33, who lives in Schwenksville, Pa., would sit at her laptop and interact with an AI-powered chatbot. For every hour of work, she was paid $20 to $40. From December to March, she made over $10,000.

The boom in AI technology has put a more sophisticated spin on a kind of gig work that doesn’t require leaving the house. The growth of large language models like the technology powering OpenAI’s ChatGPT has fueled the need for trainers like Becker, fluent English speakers who can produce quality writing.

It is not a secret that AI models learn from humans. For years, makers of AI systems like Google and OpenAI have relied on low-paid workers, typically contractors employed through other companies, to help computers visually identify subjects. (NYT has sued OpenAI and its partner, Microsoft, on claims of copyright infringement.) They might label vehicles and pedestrians for self-driving cars or identify images on photos used to train AI systems. But as AI technology has become more sophisticated, so has the job of people who must painstakingly teach it. Yesterday’s photo tagger is today’s essay writer.

There are usually two types of work for these trainers: supervised learning, where the AI learns from human-generated writing, and reinforcement learning from human feedback, where the chatbot learns from how humans rate their responses.

Companies that specialise in data curation, including the San Francisco-based start-ups Scale AI and Surge AI, hire contractors and sell their training data to bigger developers. Developers of AI models, such as the Toronto-based start-up Cohere, also recruit in-house data annotators.

It is difficult to estimate the total number of these gig workers, researchers said. But Scale AI, which hires contractors through its subsidiaries, Remotasks and Outlier, said it was common to see tens of thousands of people working on the platform at a given time.

But as with other types of gig work, the ease of flexible hours comes with its own challenges. Some workers said they never interacted with administrators behind the recruitment sites, and others had been cut off from the work with no explanation. Researchers have also raised concerns over a lack of standards, since workers typically don’t receive training on what are considered to be appropriate chatbot answers.

To become one of these contractors, workers have to pass an assessment, which includes questions like whether a social media post should be considered hateful, and why. Another one requires a more creative approach, asking contracting prospects to write a fictional short story about a green dancing octopus, set in Sam Bankman-Fried’s FTX offices on Nov. 8, 2022. (That was the day Binance, an FTX competitor, said it would buy Bankman-Fried’s company before later quickly backing out of the deal.)

Tags:    

Similar News

Editorial: Oh, to be Botswana!

Editorial: Wild things

All the Demons Are Here