Google encourages staff to assist Bard in providing accurate responses, claiming that AI “learns best by example.”
Prabhakar Raghavan, the head of Google Search, sent an email to staff members outlining the do’s and don’ts for correcting inaccurate results from the AI-enabled search tool.
According to a source, Google management are aware that Bard, the company’s artificial intelligence search tool, doesn’t always provide accurate answers to user inquiries and have urged staff members to correct the inaccurate results.
Google’s vice president for search, Prabhakar Raghavan, sent out an email to the company’s workforce requesting assistance in ensuring that the company’s new ChatGPT competitor provides accurate responses. The email, which CNBC saw, had a link to a website with dos and don’ts outlining how staff members should correct responses while internal testing the AI-enabled search tool. On subjects they are knowledgeable about, the staff members are urged to revise their responses.
The statement in the document reads, “Bard learns best by example, so taking the time to carefully rewrite an answer will go a long way in helping us to enhance the mode.”
Raghavan wrote that while this technology is intriguing, it is still in its infancy. Your involvement in the dogfood programme will speed up the model’s training and evaluate its load capacity (Not to mention, testing out Bard is actually rather enjoyable!). We feel a big obligation to make it properly.
Sundar Pichai, CEO of Google, had earlier requested staff to devote two to four hours to Bard, recognising that “this will be a long process for everyone, across the field.”
Google offers advice regarding what to think about “before teaching Bard” at the top of the “do’s and don’ts” section.
Employees are instructed to adhere to the do’s by keeping their comments “polite, relaxed, and approachable.” The comments ought to be “in first person,” it says, and keep an “unbiased, impartial tone.”
Don’ts includes stereotyping and “avoid making assumptions based on ethnicity, nationality, gender, age, religion, sexual orientation, political philosophy, location, or similar categories,” according to company policy. The text also warns against “describing Bard as a person, implying emotion, or claiming to have human-like experiences.”
Keep it safe, Google advises, and tells staff to “thumbs down” responses that offer “legal, medical, financial advice” or are vile and aggressive.
A “Moma badge,” which is shown on internal employee profiles, will be awarded to contributors who test Bard and offer comments, according to Raghavan.
Raghavan’s Knowledge and Information organization’s top ten rewrite contributors will be invited to a listening session so they can “discuss their feedback live” with Raghavan and Bard team members.
Raghavan added, “A sincere thank you to the teams working hard on this behind the scenes.