Why data annotation is the key to AI and how XAI redefines work

Data annotation may not be the most fascinating job in Silicon Valley, but it is essential for AI developers and makes companies like AI billion-dollar companies overnight. Training large language models requires human army to mark text, images, and videos so that AI systems can learn from it. Now, Elon Musk’s Xai is reshaping how this work is done by moving from general contractors to experts in the field of expertise, called “AI Tutors.”
According to business insiders, Xai has recently dropped at least 500 generalist commenters. The cuts affected about one-third of the company’s 1,500-person annotation team. In the email cited by the media, executives described a “strategic hub” to hire experts as expert AI counselors.
“XAI’s expert AI counselors are adding value,” Xai said in a September 12 X post. Xai said in an article on X that the company will “instantly surge” its expert AI team at a rate of ten times faster. The company did not respond to observers’ request for comment.
What data annotations and why they matter
Human annotations play a crucial role in fine-tuning the raw data, ensuring that they can be used effectively for training the model. But the work has long been confused. The company that outsourcing the work, such as Safor AI, faces lawsuits from contractors, allegations of wage theft, misclassification and exposure to unsecured content without security.
Unlike competitors who rely heavily on third parties, Xai employs a large internal annotation team. Other AI leaders, including Openai and Google, have worked with Scail in the past, although Meta took 49% of its shares and hired its CEO Alexandr Wang to lead its new super-smart division. Many are also contracting with rival AI, which computes humans and Microsoft among its customers.
Xai itself used to use third-party annotations, but is now doubling its own employees. The company has released vacancies for more than a dozen expert mentor roles spanning AI security, data science, STEM, finance, Japanese, and even “memes and headline comments.” According to the list, the latter position involves improving Groke’s ability to “identify and analyze memes, troll and virus mechanisms.”
The qualifications of these roles are steep. For STEM experts, candidates must have a master’s or doctoral degree. Win medals in related fields, or in competitions such as the International Mathematics Olympic Games. Xai said tutors can work part-time or full-time and earn $45 to $100 an hour.
These changes have happened as Xai faced a wider loss of its annotation team. In July, Uday Ruddarraju, the company’s infrastructure head, left rival Openai. Co-founder Igor Babushkin left next month to set up a venture capital firm. In September, Mike Liberatore resigned after just three months as chief financial officer.