[ad_1]
MOUNTAIN VIEW, Calif.—After years of Google treating visitors to its homepage to doodles, Google I/O gave me an opportunity to return the favor at its AI Sandbox demo this afternoon.Set up in a temporary structure in one of the parking lots of the Shoreline Amphitheatre here, it allowed attendees to try out some of the AI systems and tools that were discussed in the keynote that opened Google’s developer conference.My first stop invited me to challenge the Project Astra AI assistant in a doodle-recognition challenge. Handed a stylus, I drew a simple diagram on Astra’s touch screen of the District of Columbia as bisected by the Anacostia River and then saw Astra struggle with it: “It looks like a blue line with a squiggle on the left side,” it said.“Can you tell me about what you’re drawing?” it continued. After I replied with hints that it was a map and part of the US, Astra asked if I’d drawn Florida. Then I offered a riddle of a hint. “It is not one of the 50 states, but it is part of the US,” I said. That had Astra guessing Puerto Rico or Guam until I added that the subject of my map was in the continental US but lacked voting representation in Congress. That finally got Astra to recognize my representation of the nation’s capital.
The company plans on rolling out parts of Project Astra later this year through the Gemini app. The next station, “Penalty Kick Challenge,” invited visitors to take three penalty kicks into a simulated soccer net, with Google’s AI judging everybody’s form. Despite a few decades of no practice, I landed all three of my kicks around the top-left corner (OK, one bounced off the post) and then saw that the AI gave me scores of 71, 71, and 68 out of 100, factoring in sub-scores for power, accuracy, and style—whatever the last one might read.The AI had some advice for me. “That second kick was spot on, you hit the target perfectly, but across the board your power is lacking a bit,” it counseled. “Remember to lock your ankle and strike through the ball with your laces.”Then I got to order up a player card featuring a gen-AI version of one of the photos of my kicks. Being a child of the ‘80s, I had to pick “80s Retro,” while my colleague Michael Kan (who earned similar scores despite, um, having his kicks go very different places) went with “Anime Power.”
Recommended by Our Editors
We then made quick stops at the MusicFX! DJ station to try remixing various AI-generated tracks (we had to fuss with the software to get a banjo track to start playing), and an “Infinite Wonderland” demo of gen-AI art, in which I could pick a chapter from Alice’s Adventures in Wonderland and have five gen-AI routines draw art based on it. Pressing the button to have these routines redraw the art yielded scarcely different results.I wrapped up the tour at the AI Studio table. This was my opportunity to put Google’s AI—in the form of its lightweight Flash version—to work in an actual business context, writing a real-estate listing. The initial demo was interesting enough, having the AI analyze a 4:55 house-tour video before cranking out its ad copy, so I had to level up the challenge by asking the AI to rewrite the sales pitch for a doomsday prepper. It ably fielded that challenge in about 20 seconds, describing the kitchen as “Perfect for canning, dehydrating, and prepping all your survival essentials,” noting that the hot tub was “also excellent for sanitizing water, if the need arises” and endorsing the yard as “Perfect for growing your own food (when the time comes to live off the land) and keeping a low profile.”So while Google’s Gemini AI may not be ready to decipher cartographic doodling, it apparently is ready for the eco-apocalypse.
Everything Announced at Google I/O
Get Our Best Stories!
Sign up for What’s New Now to get our top stories delivered to your inbox every morning.
This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.
[ad_2]
