Today, Google opened up limited access to its AI chatbot, Bard. This is an important step in the company’s journey to taking the lead in the AI race. Initially, Bard will be available to select users in the US and UK, and you can waitlist them through the website. bard.google.com. Google notes that expanding Bard’s audience will be slow, and doesn’t specify when exactly access will open to everyone.
Like OpenAI’s ChatGPT chatbot and Microsoft’s search chatbot Bing, Google’s Bard bot invites users to ask questions about any topic in a text-entry field. Given the well-documented propensity of these bots to invent information, Google emphasizes that Bard is not a replacement for a search engine, but rather a complement to search.
“Get to know Bard, your creative and helpful assistant to expand your imagination, increase your productivity and bring your ideas to life“- wrote Google on the bot’s page. Project leaders Sissie Hsiao and Eli Collins carefully describe Bard as “an early experiment designed to help people be more productive, accelerate their ideas, and spark their curiosity“.
As with ChatGPT and Bing, there is a prominent disclaimer below the main text box warning users that “Bard may display inaccurate or offensive information that does not represent the views of Google“.
In a demo for The Verge, chatbot Bard was able to quickly and fluently respond to a number of common questions, offer mundane advice on how to get a kid bowling (“Bowling Him”), and recommend a list of popular heist movies (“The Heist”) , in Italian “bear cub” and “robbery”). Bard generates three responses for each user request, although the differences in content are minimal, and there is a “Google It” button under each response.
As expected, attempts to obtain factual information from Bard were unsuccessful. The bot was not able to answer the difficult question about the maximum load of a specific washing machine correctly, but provided three different but wrong answers. Repeating the request allowed us to get the correct information.
“This is a good example – it is obvious that the neural network is hallucinating about the maximum load., Collins said during the demonstration. — There are several numbers associated with this request, so sometimes the bot understands the context and gives the right answer, and sometimes wrong. This is one of the reasons Bard is an early experiment.“.
Also, the Google chatbot, like its competitors to ChatGPT and Bing, sometimes gives ambiguous answers, although it sometimes answers tricky questions quite adequately. In general, the chatbot shows its effectiveness in communication, and as Google will grant access to Bard to a growing number of users, this collective stress test will better identify opportunities and add limitations to the system.
It also notes that Bard is faster than the competition and appears to have the same potential capabilities as those systems. For example, it is able to generate program code. But Bard doesn’t have the well-defined Bing-like footnotes that Google says only appear when directly citing a source, such as: a news article, for example, and Bard seemed generally more reserved in his responses.