Not to be outdone by Microsoft's announcement of new AI-powered search features coming to Bing, Google held an event today to show off some of its own moves to embed AI into search.
The event presented by Prabhakar Raghavan, Google SVP for search, Assistant, ads and other products, included talk of Google's recently announced ChatGPT competitor, Bard, but the event didn't focus on it, instead running through a list of new ways Google is collapsing various forms of search - text, speech and images, for example - into a single multimodal AI-driven system.
"Search is still our biggest moonshot," Raghavan said, "and it's evasive because of constantly changing expectations and advancing technology."
Raghavan and the Googlers he had on stage talked about a variety of new AI features coming to the company's various search channels.
In terms of speech, Google said its experiments with zero-shot language learning have paid off, and it plans to add machine translation of more than two dozen languages using zero-shot translation "soon."
Google Lens was the focus of quite a bit of the presentation, with VP of search Liz Reid sharing that "search what you see" functions are coming to Lens soon, enabling users to conjure Google Assistant to identify the content of any image found on the web or texted from a friend.
Reid said multisearch with Lens, which was introduced last year and enables users to search an image with text prompts (like, "this shirt in a different color"), was now globally available, and is also providing localized results for US-based users, with plans to extend that feature globally later on.
Raghavan said that Google's new visual translation tools are also getting a new feature that will allow them to "translate the whole picture, not just the text." Raghavan said the new capability will remove text, translate it and re-add it to the image without requiring a blurred background block to hide missing image pixels.
Now, Raghavan said, text can be added back, with AI filling in the visual gaps - but only on Android devices.
Google Maps is getting a new "Immersive View" that uses AI to "fuse billions of street view and aerial images to create a 3D model of the world," said Chris Phillips, VP and GM for the Geo team.
Immersive View, which is only available in a few select cities right now with plans to roll it out to additional ones in the future, includes a time slider for viewing conditions (including forecast weather) at different times of day.
Immersive View's most impressive feature, at least as carefully curated and presented on stage, was its ability to use neural radiance fields to render 3D indoor spaces from 2D images, which Phillips demonstrated by showing it reconstructing a restaurant interior.
Google also said it's planning to start onboarding devs, creators and enterprises next month to try out a new generative language API powered by LaMDA, but said it won't have any pre-trained models - yet.
Going into today's event, the hope was that we were going to get some more details about Google's Bard, and we did - but not much.
Raghavan reiterated a lot of what was already known about Bard - like the fact that it's powered by LaMDA, presents bullet-lists of information to help users make decisions, and pulls data from the internet instead of relying on nothing but a pre-trained model with limited data set.
He also said Bard was ideal for answering "NORA" queries, or those with No One Right Answer, because the generative AI behind it can understand the big picture, explore different angles, and present multiple viewpoints.
Bard is being powered by a lightweight version of LaMDA that would allow it to be deployed and used in more places, and Raghavan said it is currently in limited release with trusted testers who are putting it through its paces before a wider release.
Unfortunately, Raghavan had nothing to say about a wider release window for Bard, and that's great news for Microsoft. Its ChatGPT-powered version of Bing, while still not generally available, at least has a public waitlist. ®
Perché siamo il Garante, e tu sei pazzo
Webinar How AI demands a new navigation
AI policy wonks slam chatty hallucination-prone model in formal complaint
Exclusive Hint: Hundreds of malicious npm and PyPI packages spotted
And it looks like educational flavor Edubuntu is returning, too
All it needs to do is detect ■■■■■■■■■■ in the ■■■■■ at ■■■■■■ when the ■■■■■■■■
Unmasking also in store for anyone who's 'posted, uploaded, downloaded or modified' tweet biz code