The Top 5 AI Announcements from Google I/O

0:00Today on the AI Breakdown, we are counting down the top 5 AI announcements from Google's IO Developer Conference. The conference was a chance for Google to get back in the game and maybe even establish a leadership position again when it comes to AI. The AI Breakdown is a daily AI news analysis show. If you're enjoying, please subscribe, leave a five-star rating or a review, or come join us on YouTube. What's going on, guys? Welcome back to the AI Breakdown. Listen, today, instead of breaking it up between the brief and the main AI Breakdown, we are doing just one show. I think that Google had so many announcements, it was dominating the news.

0:40And so we're kind of going to combine it today into this fun countdown. As I mentioned, it's the top five AI announcements from Google's IO developer conference yesterday. People were anticipating this to be very, very heavy on the AI, and I don't think that they were disappointed. In many ways, this was Google's chance for it to show that it wasn't just fully behind and losing the AI arms race, as so many have said. And so we're going to count down the top five, but we're going to actually start with an honorable mention. And for that, I'm pointing to the Adobe Firefly partnership. So basically what this partnership is, Adobe Firefly, for those of you who aren't familiar, is Adobe's full AI creative suite.

1:18So it includes their text to image tools, their video editing tools, and what they're integrating, at least initially, is effectively Adobe Firefly's text to image capabilities, but put inside BARD. So you can see here the example, make an image of a unicorn and a cake at a kid's party, the same type of prompt you'd use for mid-journey or stable diffusion or anything else. But this is happening inside Bard, right? Bard.google.com. And this comes up with the image. And so the idea here is that the experience is all integrated and people can just use this sort of generative AI right from within a single Bard experience.

1:53Now, why it gets an honorable mention is that it shows to me that Google is going to heavily leverage partnerships and effectively platformize itself in its attempt to win the AI arms race. Next, we move to the main part of our countdown number five, and I'm calling this set of announcements the AI Everywhere announcement. So it includes MedPom, MusicLM, and StudioBot. So let's start with MedPom. MedPom is a large language model from Google Research designed for the medical domain. Now, MedPom people were familiar with. MedPom 2, their second edition, was also recently announced, and this is showing really promising results.

2:30MedPOM 1 had a 67 % accuracy on the medical exam benchmark in research. MedPOM 2 has an 85.4 % accuracy. Google identifies medical question answering as one of the, quote, grand challenges for AI, and MedPOM is their answer to help in that space. Music LM is something that I am personally super excited to try. It's basically text to music. So the idea is describing a musical idea and then hearing it actually come out in a meaningful way. The example they give is ambient soft sounding music I can study to and then theoretically MusicLM will be able to just produce this. Now this is invite only testing so far so I have not had a chance to get my hands on it but you better believe that when I do I am going to use the hell out of it.

3:15Finally StudioBot is an AI powered coding assistant for Android development. Now there is a ton of coding related announcements as we'll get into, but I wanted to flag this one here as another almost throwaway relative to so much else going on. But summing it up, the reason that this set of announcements gets number five on this list is that this is really the alphabet coming out in Google's presentation, right? Google has incredible breadth when it comes to the different parts of not just technology, but our lives that it touches. It touches everything from health to music to coding and so much more than that.

3:51And when Google gives you all of these announcements about how many different domains it is inserting AI into, it's, I think, a reflection of the fact, one, that AI is going to be everywhere, but two, that Google wants to be the thing, the company that puts AI everywhere. Coming in at number four, we have one which was widely anticipated, the introduction of Palm 2. Now, Palm is Google's specific LLM, and Palm 2 is, of course, the second generation. As part of their announcement, they give a number of unique selling points to Palm 2 relative to other LLMs. First is multilinguality. They say that it spans more than 100 languages, which creates significant advances in its ability to understand, generate, translate nuanced explanations across those different language families.

4:37Number two, reasoning. It says that Palm 2 has ingested a huge number of scientific papers and web pages with mathematical expressions. So it shows a better capability in things like logic, common sense reasoning, and math. Three, they point to coding. It says that it was trained on a huge number of open source coding databases. And again, you're going to hear this throughout this video today, that coding and the use case of AI for coding is front and center of Google's plans. Lastly, an interesting little note, they write, we're already at work on Gemini, our next model created from the ground up to be multimodal, highly efficient at tool and API integrations, and built to enable future innovations like memory and planning.

5:14Gemini is still in training, but it's already exhibiting multimodal capabilities never before seen in prior models. Multimodality is obviously a huge theme. Yesterday, we talked about Meta's new image-bind open-source multimodal approach. NVIDIA is working on multimodal AI. Everyone is thinking about what comes after ChatGPT, what comes after LLMs, as the name of my video suggests yesterday, and Google is teasing that Gemini is their approach. Now, why does Palm 2 get number four? Well, of course, Palm 2 is the foundation of everything that Google's going to do in AI. How good this is as a model will dictate a lot about how good the products that use it are.

5:50And speaking of products that use Palm 2, Bard and Bard's updates and upgrades is our number three spot in terms of the top Google AI announcements. Candidly, Bard was underwhelming to a lot of people. It felt rushed in the wake of ChatGPT's success and fundamentally farther behind than it seems like Google should be. Well, they're announcing a number of updates that are trying to catch up on that front and maybe even move ahead, including things like support for 180 countries. They say that they're on track to support 40 languages. From a UI perspective, they're talking about how they're trying to make interactions with BARD more visual.

6:24Obviously, we saw that Adobe Firefly integration for text-to-image right inside BARD. But you can also see here that they're returning visuals with the search results. And in some ways, I think that what you're seeing here is Google trying to converge a new default search experience for the web, which is something we'll get into in a minute. Now, whether this makes it feel more familiar to people or whether this feels like an old out-of-date user interface is yet to be seen, but it seems to be the bet that Google is making. And if nothing else, it might be a step on the way to multimodality. Now, like in so many of these other updates, they're also really, really emphasizing the developer and coding aspect of the tool.

7:01Of course, it's relevant to recognize that the IO conference is technically a developer conference, but I don't think that they're just trying to speak to that audience. I think that they're trying to signal that coding and development is at the very core of how Google sees its approach to AI. Here they talk about a number of coding related upgrades, including sourcing citations, an export button to bring what you're doing in bar to other platforms and partners like Replit and Colab. And of course, can't miss this. They're announcing a dark theme. So this gets number three on the list because if the LLM underlying it, Palm 2, is the foundation of whatever success Google will have, BARD is the user interface on top of it that will also have a huge and dramatic impact on whether people actually want to use these tools.

7:42You could have a great model, but if it has an interface that people don't like or is hard to use or doesn't produce great results, they're simply going to go use something else. The response to BARD's update seemed to have been pretty good. Abhishek here says, ChatGPT has been dethroned. Google introduces ChatGPT competitor Barred for public, and he emphasizes the fact about how many integrations there might be right from that chat interface. Now, moving on to our number two, I think this might be my most controversial placement on this list, but for my number two, I am saying that it's Duet AI, AI for Google Workspace, and basically all of the integrations that Google has into its core work products.

8:19In many ways, these are the most boring and obvious uses of AI that Google could have pursued. It's getting AI help inside of Google Docs. It's AI support within Gmail. It's translating meeting notes into Gmail. All that sort of day-to-day, again, boring stuff that is just going to become de rigueur and routine and new uses of AI that aren't mind-blowing. They're just part of how we work in the future. Outside of Workspace, they also have a number of other announcements like this, such as Magic Editor in Google Photos, which brings the power of AI right into the photo suite, new tools for making maps more immersive.

8:53And so why is this number two? Well, I think you have to put the IO announcement in context. Just last week, we had huge, huge discussion about this leaked note from a Google researcher where they asserted that they had no moat because open source was just out developing them. A lot of people agreed with that. They thought that the innovations that were coming from open source did show just how much that community was going to shape the future of AI, but not everyone bought the argument. Sar Harabhakti here says, I have found the entire Google is falling behind in AI narrative very stupid. Today's update at I.O.

9:30is pinch-high flexing. Google will put productized generative AI across its entire ecosystem for billions of people for free this year. The bar for app layer startups just went up 10x. So of course, the point that Sar is making is that Google's moat is the fact that so many people already use Google products. And sure, they might nominally prefer the individual app AI experience of a different outside suite product, but are they really going to change from Google Docs because of that? Probably not, or at least it's a much higher barrier. So that the reason that all of these very boring integrations are my number two is that I actually think that this represents a huge part of the moat that Google does have and how they might compete going forward.

10:10But that brings us to number one, the biggest announcement from Google I.O. yesterday. What do you think it is? For me, it's got to be search. For 20 years, more than 20 years now, Google search has been at the core of the internet experience. And not just the internet experience, but the business model of the internet. The way that we interact with search is so implicit in how we engage with information now that we forget that it was a specifically designed user interface. It didn't just come fully formed. It was the product of a number of decisions that then trained us how to use it. Well, Google is now generative AI-ifying that experience.

10:50They're chat GPT-ing that experience. In fact, folks will now always have an above-the-fold generative AI part to their search results. So why is this number one? Listen, there are huge, huge implications for this. Jason Yanowitz from BlockWorks says Google revealed their new search pages yesterday. Media companies that rely on SEO for traffic are about to get crushed. Very similar to media companies that relied too heavily on social platforms for traffic. Cyrus SEO says early takeaways from Google's new AI search updates. Most answers appear to not cite sources. Links to dig deeper appear below or offside.

11:25Emphasis on shopping features. Search ads continue to appear. And like ChatGPT, you can continue conversions. So those are feature changes, but what is the result? Well, Cyrus says, honestly, it's difficult to see how this won't result in lower click-through rates for most web publishers. Now that said, some people felt that Google's announcements when it came to search weren't going far enough. Brian Rommel writes some great announcements from Google today. However, history will record the biggest error they made today is thinking the future will look like the past in tech. Brian points exactly to this search experience, saying, the idea that you will use a search engine like the late 1990s is quaint and for a company to believe is regrettable.

12:06By 2028, he says, most people will no longer use a search engine or a search bar. Your private and personal AI agent will assemble a response and use search when needed by sending agents to search engines. It was not clear to some in 2023 that the future is not like 2003. I think that is very interesting food for thought, but I have to come back to this first tweet from Ridd Design when he says, Google kind of crushed it on the AI unveiling. This was broadly the sentiment that I noticed as well, that people were impressed, that people were, that people had a sense that Google was all in on this and that whether their main competitor is meta or open AI or open source developers writ large, Google is here to compete.

12:46So that is it for the top five AI announcements from Google IO yesterday. What did I miss? What do you agree with? What do you disagree with? Debate in the comments. And if you are enjoying the AI Breakdown, please like this video, share it, subscribe to the channel. Go check out the podcast version. Go check out the newsletter, AIBreakdown.Behive.com. As always, I appreciate you listening or watching. And until tomorrow, peace.