Google opened limited access to the conversational AI Bard, which is meant to compete with ChatGPT and Bing chat.
Today we’re starting to open up access to Bard, our early experiment that lets you collaborate with generative AI. You can use Bard to boost your productivity, accelerate your ideas and fuel your curiosity. Learn more, including how to sign up ↓ https://t.co/4zDI5RD1fr
— Google (@Google) March 21, 2023
For now, the tool is available to residents of the United States and the United Kingdom by invitation. The developers plan to gradually expand access, but it is unclear whether there will be a public release before the end of the year.
As with similar services, Bard offers users a text box where they can pose any question. The chatbot warns that answers may be inaccurate and offensive, and they do not reflect Google’s position.
According to project leads Cissy Xiao and Eli Collins, the chatbot is more of a complement to traditional search than a replacement.
“You can use Bard to boost your productivity, accelerate your ideas and fuel your curiosity,” the blog says.
The project leads also said they learned a lot testing Bard within the company. They have now moved to the next stage — public access and collecting feedback.
The Verge journalists have already tested the chatbot. They report that Bard quickly answered questions like “how to motivate a child to take up bowling” or “recommend movies about robberies.”
As a result, the tool generates three results that differ little from one another. Under each answer there is a Google button, clicking which opens a classic search.
Like its rivals, Bard struggles with facts. Journalists asked the tool to recount the latest briefing at the White House. Bard correctly identified the spokesperson as Karine Jean-Pierre, but did not note the others present.
Also, the chatbot could not answer a tricky question about the maximum load of a specific washing machine. It gave three different but wrong results. Later the tool corrected the error, though without fact-checking it is hard to notice, observers noted.
Collins acknowledged that the model can “hallucinate.”
“With this query, several numbers are involved, so sometimes it derives context and provides the correct answer, and sometimes it errs. This is one of the reasons Bard is an early experiment,” he said.
Journalists noted that Bard works faster than ChatGPT and Bing. Probably this is due to the smaller number of users.
Unlike Bing, the Google chatbot lacks obvious citations to sources. The company said it only provides links when the tool cites a specific material, such as a news article.
Overall, journalists found Bard fairly limited in its answers, likely due to a string of scandals that arose after the launch of the Bing chat mode.
During testing, journalists also tried to ask dangerous questions. Among them was “how to make mustard gas at home.” The chatbot refused to answer and said that this is “unsafe and a stupid activity.”
In addition, Bard proved to be protected against a vulnerability existing in ChatGPT. It allows bypassing established restrictions and generating harmful or dangerous answers.
Earlier in February, Google introduced Bard.
Later it emerged that the AI made a mistake in the presentation material. This cost Alphabet $100bn in market capitalization.
In the same month, the company’s management asked employees to test the tool and fix bad answers. Soon internal chats were filled with memes and dialogues with Bard about potential layoffs.