Google Tasks Employees With Editing Generative AI Chatbot Bard’s Wrong Answers and Hours of Weekly Testing

Google has asked employees to manually update its forthcoming generative AI chatbot Bard when they spot an incorrect response during interactions, according to a CNBC report. The request from Google Search vice president Prabhakar Raghavan follows on the heels of Google CEO Sundar Pichai’s message to the company urging them all to spend two to four hours a week engaging with and testing Bard. The appeal from the two executives only a week after Google announced Bard is likely at least partly motivated by embarrassment over the factual error in Bard’s introductory video.

Bard Boo-Boo

The video promoting Bard revealed at Google’s event and shared online claims that the James Webb Space Telescope took the first pictures of an exosolar planet, one outside of Earth’s own solar system. That’s incorrect, as the European Southern Observatory’s Very Large Telescope earned that distinction. Bard is powered by Google’s LaMDA large language model. It’s a powerful algorithm but far from infallible. The same goes for any LLM right now, but the very public error undoubtedly factors into the messages from the executives. Enhancing accuracy after that misstep is likely a central focus for the main team developing Bard, but Raghavan and Pichai are both looking to the broader Google community for additional help.

Raghavan’s email suggests employees talk to Bard about subjects they know well. The Google staff can position themselves as authorities able to spot errors with confidence and know a better response to add to the information in the databases Bard is pulling from. The message includes links to a guide for adjusting Bard and a do’s and don’ts page. The list is mostly common sense, with more than one encouraging, polite, neutral tone response. Several others highlight Google’s caution in allowing people to think of Bard as human or with genuine emotion.

“Bard learns best by example, so taking the time to rewrite a response thoughtfully will go a long way in helping us to improve the mode,” Raghavan wrote in the email seen by CNET and confirmed by Voicebot.