Google AI chat robot Bard auditor said he was under a lot of pressure and had only 3 minutes to review the answers.

CTOnews.com, July 13 (Xinhua)-- Google's artificial intelligence chat robot Bard is trained under pressure by thousands of contract workers (Contractor) who have only three minutes to review Bard-generated responses, Bloomberg reported.

The contractors, reportedly from companies such as Appen and Accenture, received minimal training at an hourly rate of $14 (CTOnews.com Note: currently about Rmb100). Their job is to assess whether Bard's answer is reliable and accurate.

Although chatbots like Bard and ChatGPT rely on underlying large language models to generate answers, humans are also involved in the review process to ensure that they are reliable and accurate. However, according to internal documents and six contractors, the human workload of reviewing Bard responses is becoming increasingly large and complex, according to Bloomberg.

"as far as the current situation is concerned, people feel scared, stressed, underpaid and don't know what happened," said a contractor. "this culture of fear is not conducive to our quality and teamwork."

One of the tasks of the contractor is to evaluate its "usefulness" based on its answers and rate it on a range from "useless" to "very useful".

"giving people access to high-quality information is at the core of our mission," a Google spokesman said in a statement. We have done a lot of work to build our artificial intelligence products responsibly, including the rigorous testing, training and feedback processes that we have been improving over the years to emphasize facts and reduce bias. Manual evaluation, including individuals from inside and outside Google, is one of the many ways we use to improve our products. "

"scores do not directly affect the output of our models, and they are by no means the only way for us to improve accuracy," the spokesman said. "