Google I/O 2024 live updates: The latest on Gemini AI, Android 15 and more

Thanks for joining us. We're off to try and get a closer look at some of the things Google announced today at I/O.

Oh, and that Sundar bumped that final AI count up one more time to 121.

Who wants to bet if Google will top that next year?

Frankly I'm just glad it didn't last any longer than that. While Sam and Karissa go and check out whatever demos might be available at I/O 2024, come on over and join myself and senior reviewer Devindra Hardawar on Engadget's YouTube channel. I'm sure you have thoughts and feelings to share on everything Google just announced!

And that's a wrap on the keynote, it was *just* under 2 hours of all things Gemini and AI.

120 mentions of "AI" in this keynote, according to Sundar (who just came back onstage) and Gemini. I actually would have guessed a bit higher.

Google, like most of its AI peers, plans to use watermarking to increase transparency around AI-generated content with SynthID. Those watermarks will be expanding to AI-generated video (remember those Veo demos from earlier?) and text, which is especially interesting because AI-generated text is so much more prevalent (and in some ways harder to detect than images or video).

SynthID is a tool that Google is making so that it's easy detect AI-generated content. But it also sounds like something from the Blade Runner universe that people can use to detect replicants.

We're getting into "building AI responsibly" portion, with a look at Google's approach to red-teaming, the process for looking for threats, "problematic" outputs and other issues that might cause problems for Google (and maybe the rest of us). After the fiasco with Gemini's image generator earlier this year, I'm actually surprised they waited this long to bring all this up.

So it appears we've reached the developer-focused section of the I/O keynote. Karissa and I were just talking about how for non-developers, it can be hard to contextualize the amounts of tokens available in various Gemini plans.

We're getting a deeper dive into Gemini Flash, which we first heard about at the top of the keynote. Flash is the "lighter weight" Gemini model that has lower latency and is more efficient. It sounds like this is meant for developers and organizations that don't need the full power of Gemini 1.5 Pro but still want to take advantage of multimodal capabilities.

Yea, I think the scam alert feature is really nice, because it can give you a gut check in real-time when you might be feeling that a call feels a little sus.

Josh Woodward is back to talk about various Gemini models available to developers.

Google has another solution to the never-ending scam calls. If you pick up, and it detects that the call is "suspicious" or a likely scam, it can surface a big alert right on your phone, potentially saving you from buying thousands of dollars of gift cards or transferring funds to a rogue account. A lot of those scams seem obvious, but people still fall for them and this might help.

Like, will we ever see the day where you're allowed to bring a phone to take the SATs? Because if not, you're potentially giving students an aid they might not being able to use during important evaluations, which might end up being a disadvantage.

Karissa, I think Circle to Search supporting formulas is weird because for older types like us (or at least me), it feels like cheating.

I hate searching for something and getting a 5-minute YouTube video as a result so being able to "ask this video" instead is something I could definitely see myself using. I wonder if you still have to watch the pre-roll?

I understand why it would be helpful for Google to help with homework questions but there's something that still feels a bit off about encouraging students and parents to punt homework help over to Google's AI.

OK, sidenote: They showed a photo of someone searching for Fly by Jing sauce. That stuff is great.

We've finally started the Android portion of the keynote. Wow, Gemini is "becoming the new AI assistant on Android" so I guess Google Assistant is dead?

Oh no, we've hit the musical interlude for Google I/O.

I agree. I experimented with using chatGPT and other AI tools to plan an itinerary last summer and it was a bit ... rough. If anything, it was more of a starting place than an actual solution, especially because several of its suggestions were straight-up hallucinations! Maybe Gemini will do better,

Now Sissie is talking about planning a trip with Gemini, and I feel like vacations are too expensive to trust an AI to plan. It's going to take some time for people to feel comfortable with letting AI take the wheel like this.

Gemini's app is getting a new "Live" feature that brings in the Project Astra tech we saw early in the keynote. It can speak and respond to voice commands and use your phone's camera for input.

The tech might be there, but I really don't think humans are ready for AI co-workers. Can you even imagine people who struggle with email trying to talk to smart chatbots in Slack?

Chip is essentially an AI coworker.. it can analyze shared documents, schedule meetings and track projects across your company's workspace. It sounds like organizations will be able to customize their own versions of Chip to suit their needs.

We're officially one hour into this keynote, by the way, and still no love for Android. I know there will be plenty of time for Android updates later on, but it says a lot about how far Google's priorities have shifted, it wasn't that long ago that Android was the star of I/O. Now, it feels almost like an afterthought.

Though I guess that's sort of skipping ahead of where AI is at right now.

But I have to wonder if this role might be better served by an AI agent that would do these same things with potentially even less manual input.

Yea, I'm not going to lie, the ability of AI's to summarize things is nice, but it feels like the lowest rung of its capabilities. So it's good to see how things are evolving with contextual replies and carrying over data between various apps like Gmail, Calendar, etc.

Gemini is going to power some interesting new features in the Gmail app, including summaries of long email threads. You'll also be able to chat with Gemini directly to find details from throughout your inbox. There are also new reply suggestions that can understand the context of your emails for (hopefully) more useful text suggestions. I've never been particularly impressed with Gmail's auto-suggestions but maybe Gemini will make them actually useful.

Yea, making all this work in the real world isn't something you can just snap your fingers and voila. But down the line it almost makes me wonder if humanity's role going forward might be to just fact-check whatever army of AI agents you have at your disposal.

Okay, this actually seems pretty useful. Search is getting video recognition capabilities so you can search based on the contents of a video, which could be particularly helpful when you need to fix something or answer a question you can't easily describe. Google will be able to understand what's in the clip and (hopefully) surface relevant results.