Google just launched Bard, its answer to ChatGPT and wants you to improve it
Google has a lot to do with this launch. Microsoft has teamed up with OpenAI to make an aggressive play for Google’s top spot in search. Meanwhile, Google blundered early on when it first tried to answer. In a teaser clip for Bard that the company released in February, the chatbot was shown making a factual error. Google’s value plummeted by $100 billion overnight.
Google won’t share many details about how Bard works: the big language models, the technology behind this wave of chatbots, have become valuable intellectual property. But he will say that Bard is built on a new version of LaMDA, Google’s flagship large language model. Google says it will update Bard as the underlying technology improves. Like ChatGPT and GPT-4, Bard is honed using reinforcement learning from human feedback, a technique that trains a large language model to yield more useful and less toxic responses.
Google has been working on Bard for a few months behind closed doors but says it’s still an experiment. The company is now making the chatbot available to people in the US and UK who sign up for a waitlist. These early adopters will help test and improve the technology. “We will receive feedback from users and will increase it over time based on that feedback,” Google says.vice-president for research, Zoubin Ghahramani. “We are aware of everything that can go wrong with large language models.”
But Margaret Mitchell, chief ethics scientist at AI startup Hugging Face and former co-lead of Google’s AI ethics team, is skeptical of that framing. Google has been working on LaMDA for years, she says, and she thinks pitching Bard as an experiment “is a PR trick that big companies use to reach millions of customers while evading liability if something goes wrong. “.
Google wants users to think of Bard as a Google Search sidekick, not a replacement. A button below Bard’s chat widget says “Google It.” The idea is to entice users to head to Google search to check out Bard’s answers or learn more. “That’s one of the things that helps us compensate for the limitations of technology,” Krawczyk says.
“We really want to encourage people to explore other places, to kind of confirm things if they’re not sure,” Ghahramani says.
This recognition of Bard’s flaws shaped the design of the chatbot in other ways as well. Users can only interact with Bard a handful of times during a given session. Indeed, the more large language models engage in a single conversation, the more likely they are to go off the rails. Many of the weirdest Bing Chat responses people have shared online have come at the end of endless conversations, for example.
Google won’t confirm what the chat limit will be for launch, but it will be set quite low for the initial release and adjusted based on user feedback.
Google also plays it safe in terms of content. Users will not be able to request sexually explicit, illegal or harmful material (as judged by Google) or personal information. In my demo, Bard didn’t give me any advice on how to make a Molotov cocktail. This is the standard for this generation of chatbot. But it also wouldn’t provide medical information, such as how to spot signs of cancer. “Bard is not a doctor. He’s not going to give medical advice,” Krawczyk explains.
Perhaps the biggest difference between Bard and ChatGPT is that Bard produces three versions of each response, which Google calls “drafts”. Users can click between them and choose the answer they prefer, or mix and match them. The goal is to remind people that Bard cannot generate perfect answers. “There’s a sense of authority when you only see one example,” Krawczyk says. “And we know there are limits to factuality.”
Leave a Reply