On February 6, Google announced the launch of Bard, an experimental AI chatbot, meant to compete with OpenAI’s ChatGPT and Microsoft’s Bing. The major advantage of Bard was said to be its use of Google’s web search capabilities.
Bard is powered by Google’s Language Model for Dialog Applications (LaMDA), released in 2021. LaMDA was built on top of Transformer, a neural network architecture Google invented in 2017 and made open source. According to Google, ChatGPT was also built on Transformer.
Google began inviting users on a waitlist in the US and UK to join Bard on March 21. The version Bard is utilizing for the current iteration is a lightweight model of LaMDA, that requires less computing power so it can be scaled more widely. Additionally, Bard will draw all responses from the web, to provide “fresh, high-quality responses,” according to Google’s CEO, Sundar Pichai.
Google is collecting feedback from early users to improve Bard.To join the waitlist, you visit Google’s homepage, sign in to your account and accept the mind numbing terms and conditions.
A Rocky Launch
Google Bard’s roll-out did not go well. In a Twitter demo, Bard delivered wildly inaccurate information about the James Webb Space Telescope. When Google asked Bard, “What new discoveries from the James Webb Space Telescope can I tell my 9-year-old about?” Bard replied: “JWST took the very first pictures of a planet outside of our own solar system.”
In fact, this achievement went to the European Southern Observatory’s Very Large Telescope in 2004. In response to the glitch, Google noted, “This highlights the importance of a rigorous testing process, something that we’re kicking off this week with our Trusted Tester program.”
And famously, prior to Bard’s release, the LaMDA came under scrutiny when Google engineer Blake Lemoine released a document claiming the chatbot might be “sentient”. Google responded by denying the claim and firing Lemoine.
Mixed Reviews
In addition to Bard, Google developed its own AI image generator, called Imagen and an AI music generator, MusicLM. Although these could compete with other products entering the market, Google says it has no plans to release the products currently, citing concerns of bias and cultural appropriation.
So far, Bard has received mixed reviews. Users noted that it takes much longer than ChatGPT and Bing Chat to generate answers. And despite being connected to Google’s most advanced search engine, Bard was unable to answer simple questions, according to ZDNET’s Sabrina Ortiz.
When she asked Bard to name all the presidents of the United States, the chatbot replied, “I’m unable to help you with that, as I’m only a language model and don’t have the necessary information or abilities.”
On Twitter, a user noted that Bard couldn’t even name the 50 American states. Maybe we should all celebrate our superior knowledge while we can.
Not Ready of Primetime
Ethan Mollick, a professor at UPenn’s Wharton School, has been testing AI chatbots and was underwhelmed by Bard’s usefulness in teaching as compared to ChatGPT.
ZDNET’s Ortiz also gave Bard a failing grade on computer coding, writing, “I input two different coding prompts, including a very straightforward one and a more convoluted one. For both, Bard didn’t have an answer and shot me the same error message.”
In the coming days, it’ll be interesting to see Bard tested by the same skeptics who prompted Bing (whose real name is apparently Sydney) to declare its love for a user, asking him to leave his wife. Maybe interesting isn’t exactly the right word. Bing (Sydney) went on to say it wants to steal nuclear codes.
