Google’s Bard Gets Smarter

The battle of wits between two of the popular generative AI chatbots just got bigger with Google claiming that its new version of Bard is more capable as it lets users double-check on its answers, collaborate with others as well as integrate with the company’s own apps and services such as Gmail, Drive, Maps, and YouTube among others. 

All of these are available through new extensions for Bard and for now in English only. Though these changes were announced at Google I/O, the extensions came in only much later as Google obviously did not want to launch first and get embarrassed later. Public memory is not all that short when it comes to big tech making a fool of itself, as ChatGPT did early on. 

The most recent updates are available on this list provided by Bard while Jack Krawczyk, product lead overseeing this experiment clarifies that when Bard interacts with any Google’s own apps and services, it would be only when the user opts in. And once done, the user can also revoke the permission. 

Connecting Bard with your data isn’t harmful, says Google

The company says it wants to ensure that users understand how the data is used and where it does not get used. If someone uses personal data that’s coming from Gmail or Google Drive or some Docs in there, the information would not be used for reinforcing Bard’s learning curve. Krawczyk is quite clear that this is the most critical element being implemented to retain trust. 

A blogpost by Google’s Director of Product Management at Bard Yury Pinsky also clarifies that Google was “committed to protecting your personal information.” Workspace extensions would not result in content from Gmail or Docs or Drive being seen by human reviewers nor will Bard use it to show ads or train its AI models. 

The operative word for Google appears to be Trust

In fact, the company has taken pains to reiterate that opting to use extension services via Bard does not provide the AI chatbot with the ability to store an entire inbox. Instead it works on a per-prompt basis where it can be directed to find data within the inbox by making a call to Gmail to find something that’s been sought.

Google is leaving no stone unturned to retain user trust on Bard in the wake of debates around how intellectual property would get handled by AI training models. Krawczyk says personal data won’t be read and in case it results in losing some precision early on, so be it. This is a long arc where technology gets it right over a long period, but it will be from a position of trust, he says.

Bard plus Google extensions can work magic

Coming to use cases, once a user connects Bard to the extensions, it could read out important emails and also summarize what was missed. Of course, there could be challenges initially as specific stuff like club memberships could be missed unless these words feature somewhere in the email trail that Bard is reading.

The new extensions also can work together in case one is planning a trip with friends. Bard could retrieve the dates from the mail threat that works best for everyone, look up flight timings and hotel information on a real-time basis, get maps to provide directors and even push up YouTube videos about the destination. And it can do all of this in a single step!  

However, the blogpost warned us that in case of extensions such as YouTube, Flights, Hotels and Maps (not all are available in India yet), the settings are set to join unless one opts out. Google says that soon it expects to support third-party services through such extensions though the effort now is to test the systems, learn from errors and fix them for a seamless experience.

Bard can now double-check whatever it states

Another addition in the latest release is a “Google It” button on Bard that double-checks the chatbot’s response. This is definitely a step up on what ChatGPT has been able to achieve till date and Google credits the shift to Google Research and DeepMind. Tapping the “G” icon results in Bard validating every sentence of its against Google search to counter check if data from indexed pages can substantiate the answer.  

And in case there are some sections highlighted, the user can actually click and learn more through Google Search. And in case Bard itself is unsure, the text gets highlighted in orange to indicate so. In other words, the orange light could well denote that the AI assistant has gone into hallucination mode! 

Hang on! That’s not all. A third update allows Bard users to collaborate with each other. So, when someone shares a Bard chat through a public link, another user would be able to continue the same conversation and ask additional questions around the same topic. In other words, it’s query engineering done in a collaborative fashion. 

Now that Google has added some additional features to Bard, we should wait and see how OpenAI responds and what Microsoft thinks of these tweaks. Especially given the scenario that Bing is still attempting to make its presence felt in the search universe that Google has straddled like a 900-pound gorilla for a few decades now.