Revolutionizing AI: Google's Vertex AI Unveils Massive Upgrades

0:00I think the first thing to talk about here is that really is the adoption of AI continues to gain traction in the business world. Google Cloud is really kind of intensifying its efforts to remain competitive. The space is evolving very rapidly. Microsoft and Amazon are diving in headfirst. And if Google Cloud wants to keep some of the gains it has made, essentially compared to Amazon over the last couple of years, they're going to have to hit AI pretty hard. For those that don't know, essentially, if you looked at the cloud landscape, you know, five years ago, AWS completely dominated it. And then ever since then, I would say they've been the major dominant player.

0:36But over the last three or four years, Microsoft and Google have started to kind of edge away some of that kind of massive gain and kind of that massive market share that AWS has had. And I believe that Microsoft with Azure and Google with cloud right now really feel like now is the time to take away a little bit of AWS's thunder, which has been a massive cash cow for Amazon, and start getting some of that revenue by integrating AI. So a recent survey actually was done by Fortune and Deloitte as well as McKinsey, and they all underscored this massive surge in AI experimentation by organizations.

1:12So over half of global CEOs are currently tinkering with AI in some form, and a third are using generative AI in at least one business function. So I think really I'm keen to kind of capitalize on this growing interest. Google used its annual Cloud Next conference, which just happened. And so we're getting a ton of different news about AI out of Google. As you know, these days when Google does a conference, everything's about AI. And they essentially have unveiled updates to Vertex AI, which is its platform designed to simplify the development, training, and deployment of machine learning models.

1:48So Jun Yang, who's the VP of Cloud AI and Industry Solutions at Google, emphasized the company's, quote, open ecosystem approach, which essentially aims to give customers a range of choices by incorporating third-party models from startups like Anthropic and Meta, right? This is not just, you know, throwing an API to exclusively open AI on there. So Vertex AI now has improved AI models for text, image, and code generation. Google claims a 25 % quality improvement in its upgraded Codi code-generated model for major supported languages. So although specifics are lacking, they didn't really go into exactly backing up that claim, which makes it a little murky.

2:29um imgen which is google's image generation model now supports style tuning essentially a letting customers create brand aligned images using a minimal number of reference pictures so furthermore the platform's palm 2 language model has broadened its language understanding covering 38 languages in its general uh availability and over a hundred in its preview that is now you know they're letting people use so the models expanded 32 000 token context windows allows it to consider up to 25 000 words or around 80 pages of double spaced text i don't know why they you know they were like kind of double spaced okay so 40 pages of single space text whatever in any case um it's able to do that before generating additional text this is actually really impressive i love to see this in models and it's kind of funny because right now i think there's a lot of models distinguishing themselves by this or like how many tokens essentially they can take as an input, where inevitably and eventually every model is just going to probably have unlimited and you just probably have to pay for it in some way, like one way, shape or form.

3:33But right now it feels like there definitely are some limits. I've talked about this on other podcasts, so I won't go into it, but it makes a big difference to be able to have, you know, 40 pages. You give it a transcript, a book, a report, and have it go through that and give you some insight. So Nenshad, who is a product leader at Vertex AI, stated that the choice of a 32 ,000 token window kind of strikes a balance between new capabilities and competitive price to performance ratios, right? The problem is if the token windows are a lot bigger than that, it becomes just incredibly expensive. So it would appear at the moment, we're still trying to figure out this whole balancing act thing.

4:07But for those who need even more extensive context windows, Google has integrated third party models like Anthropics Cloud 2 into Vertex AI's Model Garden, and is essentially offering a collection of pre-built models that can be customized to an enterprise's needs. So this move is seen as a direct challenge to Amazon Bedrock, AWS's equivalent platform, particularly given Bedrock's sort of rocky launch that they have seen recently. So in terms of utility, Google is also introducing extensions and data connectors to Vertex AI, essentially providing similar functionalities to OpenAI and Microsoft's kind of AI model plugins.

4:45and these tools are going to allow developers to connect models to real-time data or third-party apps such as customer relation management systems all that kind of stuff has really started building out some of these integrations so i think despite these advancements the platform has not really fully addressed a lot of lingering concerns legal issues surrounding generative ai there's a lot of questions about copyright status of the content generated and whether you know customers can actually claim ownership. Google insists that it conducts rigorous, quote, data governance reviews, but it's sort of unambiguous still, so it's kind of hard to know exactly what that means.

5:22I would say two other notable developments are Vertex AI Search and Vertex AI Conversation, which are designed to facilitate the development of AI-powered chatbots and search engines. So these services can be combined with model extensions and data connectors, as well as a lot of different grounding capabilities to make the AI outputs more reliable and kind of rooted in authoritative data. But even though Google has made efforts to evaluate every API call to Vertex hosted, you know, generative models for quote safety attributes, challenges definitely still remain. So generative AI models are still prone to generate, you know, hallucinate or generate some fake information.

6:05And there are still ongoing concerns about different ideologies that are embedded into AI models that could be problematic. So as some people are putting it, while the current tools may not solve all the issues with generative models, I think they're a step in the right direction. But all in all, I think Google's update to Vertex AI reflects kind of the broader industry trend towards deeper integration of AI into various business functions. I think it's developing essentially a narrative that is pretty undoubtedly worth keeping an eye on as Google Cloud and its competitors kind of are in this big battle right now to see who is going to get the most market share in this space.

6:45So this is an area we'll continue to follow in the future.