AI "training" profession

ChatGPT logo at its office in Washington, DC, US. Illustration photo: AFP/TTXVN

This job helped him earn $15/hour, higher than the minimum wage of people in this city.

Though it has fallen out of the spotlight as ChatGPT has grown in popularity, Savreux and his contractors have spent countless hours over the years “training” OpenAI’s systems to respond more accurately. Their work fills a pressing and ongoing need for AI companies to provide input data—text, brand names, and other information—to train their AI engines.

Savreux said that while this is not a job that will bring fame or fortune, and might even be considered trivial, it is essential. “You can design any neural network or hire any researcher you want, but without someone collecting, curating, and labeling the data, you don’t have ChatGPT. You don’t have anything,” Savreux said.

Sonam Jindal, head of the AI, labor and economy program at the Partnership on AI, said there is a lot of discussion around AI, but almost none of it addresses the fact that the technology still relies heavily on human labor.

For decades, the tech industry relied on the labor of thousands of low-skilled, low-wage workers to build the tech empires we see today. Their work was often temporary and largely invisible once the company’s technology became mainstream, with the executives and researchers taking all the credit. This model has continued into the AI ​​space.

In a 2021 report, the Partnership on AI noted that demand for “data enrichment” jobs is on the rise. Despite this, the report called on tech companies to develop voluntary guidelines, commit to fair pay for contract workers, and make other improvements. So far, Deepmind, Google’s AI subsidiary, is the only tech company to publicly commit to these principles. Jindal said many people have recognized the importance of AI training. This is a new job created by AI, and the contributions of workers in this industry should also be respected and appreciated.

While there are no specific statistics on the number of contract workers working for AI companies, it is safe to say that this is an increasingly common job around the world. According to online news agency Semafor, OpenAI has hired about 1,000 people in Eastern Europe and Latin America to label data for malicious content or train the company’s software on computer engineering tasks. OpenAI is still a small company, with only about 375 employees as of January 2023. However, this number is not enough to reflect the entire scale of the company’s operations, as it does not include contract workers.

In addition, generating data to train AI models is not a simple task and is complex enough to attract would-be AI entrepreneurs. Jatin Kumar, a resident of Texas, said he worked as a contract AI trainer for a year after graduating from college with a degree in computer science. This was an opportunity for Kumar to explore new-generation AI technology while developing his own technology company, Bonsai, which makes software to help pay hospital bills.

Kumar said he worked with about 100 other contractors on tasks such as creating training data, correcting answers, and fine-tuning the model by providing feedback on the answers. They also took chats that users flagged as inappropriate, sorted them into relevant errors, and used them to train other AI models. Kumar said that while he started this job as a way to help OpenAI and learn about existing technology, he eventually came to see it as his true passion.

According to VNA