Google has loads riding on this launch. Microsoft partnered with OpenAI to make an aggressive play for Google’s top spot in search. Meanwhile, Google blundered straight out of the gate when it first tried to reply. In a teaser clip for Bard that the corporate put out in February, the chatbot was shown making a factual error. Google’s value fell by $100 billion overnight.
Google won’t share many details about how Bard works: large language models, the technology behind this wave of chatbots, have turn into useful IP. But it’ll say that Bard is built on top of a new edition of LaMDA, Google’s flagship large language model. Google says it’ll update Bard because the underlying tech improves. Like ChatGPT and GPT-4, Bard is fine-tuned using reinforcement learning from human feedback, a method that trains a big language model to present more useful and fewer toxic responses.
Google has been working on Bard for a couple of months behind closed doors but says that it’s still an experiment. The corporate is now making the chatbot available without spending a dime to people within the US and the UK who enroll to a waitlist. These early users will help test and improve the technology. “We’ll get user feedback, and we’ll ramp it up over time based on that feedback,” says Google’s vp of research, Zoubin Ghahramani. “We’re mindful of all of the things that may go unsuitable 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 this framing. Google has been working on LaMDA for years, she says, and she or he thinks pitching Bard as an experiment “is a PR trick that larger corporations use to achieve hundreds of thousands of consumers while also removing themselves from accountability if anything goes unsuitable.”
Google wants users to think about Bard as a sidekick to Google Search, not a alternative. A button that sits below Bard’s chat widget says “Google It.” The concept is to nudge users to move to Google Search to examine Bard’s answers or discover more. “It’s one in every of the things that help us offset limitations of the technology,” says Krawczyk.
“We really need to encourage people to really explore other places, form of confirm things in the event that they’re undecided,” says Ghahramani.
This acknowledgement of Bard’s flaws has shaped the chatbot’s design in other ways, too. Users can interact with Bard only a handful of times in any given session. It’s because the longer large language models engage in a single conversation, the more likely they’re to go off the rails. Most of the weirder responses from Bing Chat that individuals have shared online emerged at the tip of drawn-out exchanges, for instance.
Google won’t confirm what the conversation limit can be for launch, but it’ll be set quite low for the initial release and adjusted depending on user feedback.
Google can also be playing it protected by way of content. Users is not going to have the opportunity to ask for sexually explicit, illegal, or harmful material (as judged by Google) or personal information. In my demo, Bard wouldn’t give me recommendations on find out how to make a Molotov cocktail. That’s standard for this generation of chatbot. However it would also not provide any medical information, resembling find out how to spot signs of cancer. “Bard just isn’t a physician. It’s not going to present medical advice,” says Krawczyk.
Perhaps the largest difference between Bard and ChatGPT is that Bard produces three versions of each response, which Google calls “drafts.” Users can click between them and pick the response they like, or mix and match between them. The aim is to remind people who Bard cannot generate perfect answers. “There’s the sense of authoritativeness while you only see one example,” says Krawczyk. “And we all know there are limitations around factuality.”