Google told its employees to train a competing chatbot (ChatGPT) to give the correct answers after wrong answers led to widespread criticism and a $100 billion drop in market value.
Executives at the US tech giant realized that Bard wasn't always accurate when answering questions, and now it looks like they're going to ask employees to correct wrong answers.
Prabhakar Raghavan, vice president of research at Google, asked employees in an email on Wednesday to help the company ensure Bard provided the correct answers and encouraged employees to rewrite answers based on topics they know.
“(The bad) learn best by example, so taking the time to rewrite thoughtful responses will go a long way toward improving our situation,” the document reads.
Raghavan's comments echo what Google CEO Sundar Pichai said yesterday (Wednesday) when he asked employees to spend two to four hours with Bard, admitting that "it's been a long journey for everyone."
"It's an exciting technology, but it's still in its infancy," said Raghavan. We feel very responsible and your participation in the pilot program will help accelerate model training and viability testing. "
Google announced its cool chatbot last week, but a series of glitches in the bot caused the company's shares to drop nearly 9%. Staff criticized Pichai's accident, calling the robot's launch "sloppy," "a failure," and "ridiculously short-sighted."
When trying to troubleshoot AI, business leaders rely on human knowledge, and materials sent to employees include a "do's and don'ts" section where Google gives advice to consider before training (bad).
For the "Do" section, Google asked employees to keep their responses "polite, casual, user-friendly," "in the first person," and a "neutral and reserved tone." In the "Prohibited Matters" section, Google prohibits its employees from making negative assumptions and assumptions based on race, national origin, gender, age, sexual orientation, political ideology, or location.
Google also prohibits employees from describing Cool as a person, expressing feelings, or pretending to have human experiences. He asked them if they disliked the bots' responses, which included "legal, medical and financial advice", or if they found them offensive and insulting.
As an incentive for employees to attend Bard training and provide feedback, Raghavan said employees receive a Moma badge, which displays an internal employee profile.