Sundar Pichai onstage at Google I/O 2023.
Google is supercharging Bard by giving it access to its most popular apps.It’s a big leap forward as the company continues to chase OpenAI.Bard is now also set to highlight parts of responses that may be hallucinations.
Google is supercharging its Bard AI chatbot by giving it access to the company’s most popular apps, such as Gmail, Maps, and YouTube.
The new abilities should make Bard significantly more useful, underscoring Google’s advantage in tapping hugely popular existing products as it continues to race against rivals such as OpenAI.
Bard can now retrieve information from YouTube, Maps, and Google’s Flights and Shopping search features. Also, if granted permission, it can access information from users’ personal content from Gmail, Docs, and Drive to summarize documents.
Google says you can ask Bard to find and bring together information from relevant emails. For example, you could ask it to pull together emails related to a coming vacation and summarize that information.
It’s set to combine extensions, too. You could ask Bard to find a document you previously created of your bucket-list restaurants and then show their locations on a map. Or you could upload a photo to Lens and ask it to find relevant YouTube videos.
The added features are the result of a new version of Google’s PaLM 2 large language model. Sissie Hsiao, a Google vice president who is the general manager of Assistant and Bard, says the new version has better fine-tuning.
Hsiao told Insider the information taken from apps such as Gmail or Docs would not be used for advertising or for training Google’s AI models.
The new abilities are a big step forward for Google’s chatbot and serve as a reminder of the advantage companies such as Google have in the AI race. Google has nine products with over 1 billion users that it’s eagerly weaving AI into.
“This is really the first step in taking an LLM and not just as a text model, but actually bringing it into what we call agentive capabilities,” Hsiao said, likening the updated Bard to a “personal assistant.”
Google has been working fast to iterate Bard after the company was surprised by the arrival of ChatGPT last year. Bard had a tough start, but Google has moved quickly on improving the chatbot’s capabilities in recent months.
Still, Bard is at risk of spitting out misinformation. As part of the new update, Google says it will highlight responses in green where Bard’s information is corroborated with a webpage; users can click through to that webpage if they want. Parts of the response that Google cannot corroborate will be highlighted yellow, it says. This could help users identify so-called hallucinations, or false details that AI chatbots sometimes invent.
Google has already rolled out AI-assistant features to its Workspace apps and has clarified how it plans to keep weaving AI into its most popular products.
Google is also busy at work on Gemini, a new large language model that’s set to take the capabilities of Bard and other products further. It may not be far away: Gemini is already in the hands of some developers, The Information reported.
Got a tip? You can reach this reporter via encrypted email (hlangley@protonmail.com) or the encrypted messaging apps Signal/Telegram (+1 628-228-1836). No PR pitches, please.