GOOGLE CLOUD NEXT, SAN FRANCISCO – These kinds of events always come with a massive information dump that can make it hard to know what news is worth focusing on. That was especially the case Tuesday when it seemed like there were a million artificial intelligence (AI) baubles being dangled from all angles. Lucky for you, Silverliners, we’re taking one for the team and boiling everything down to the nuts and bolts.
The takeaways? Vertex AI is expanding, Duet AI is tackling refactoring and database migration, and AI advances in Workspace mean you may never have to attend a virtual meeting again (don’t tell your boss).
Vertex AI
Nenshad Bardoliwalla, the product lead for Vertex AI at Google Cloud, told Silverlinings that Vertex AI is Google Cloud’s machine learning platform that provides the foundational capabilities to build machine learning systems. So, for instance, it can be used to build machine learning models from scratch and fine tune or customize existing models.
The models built in Vertex AI live in what the company calls a “model garden” – you can think of it kind of like an app store. You go in, choose the model you want and —boom! — you’re on your way.
Keep up with the all the news from Google Cloud Next 2023 with our dedicated news hub here.
Vertex AI made its general availability debut in 2021 and initially only tackled a handful of use cases with a limited number of models. But during a keynote speech, Google Cloud CEO Thomas Kurian said it is expanding the number of models available to over 100, with newcomers including Meta’s Llama 2 and Code Llama models, as well as Anthropic’s forthcoming Claude 2 and other third-party offerings.
Several of Google’s own Vertex AI models got significant updates as well. For instance, its text-focused AI model, PaLM 2, can now analyze documents that contain up to 32,000 tokens. That’s about 80 pages or so of information. Additionally, its picture-generating model Imagen now includes image editing capabilities, digital watermarking and the ability to train in an enterprise’s unique style characteristics using just 10 images.
Executives said customers like Vodafone, Estée Lauder and UKG have swooped in to build projects on the platform. Already, Vertex AI is home to more than 150 generative AI (GenAI) projects. On a pre-briefing call with press, executives said Google Cloud’s GenAI customer count has grown 15x over the last quarter alone.
Kurian stressed that Google is not hoovering up all the data that enterprises are plugging into Vertex AI.
“Vertex gives you full control and segregation of your data, your code and your intellectual property with zero data leakage,” he stated, adding that the word data here encompasses inputs, metadata, outputs and human feedback. “You are not exposing that data to general foundation models, Google’s consumer services, other the customers or the internet.”
Duet AI
Born out of Vertex AI, Duet AI is basically a model on steroids. That is, it provides a model, configuration characteristics and application logic to address a specific set of applications and processes around code and chat assistance, Bardoliwalla told Silverlinings.
Google unveiled Duet AI at its I/O conference in May and is now expanding its capabilities across Google Cloud. The big news here for anyone with legacy mainframe or other IT infrastructure is that Duet AI can now help with database migration and application modernization using natural language prompts. That latter process is also known as refactoring.
Typically, writing code would require specialized knowledge. But Duet AI aims to remove that barrier and can convert old code to new cloud code with contextual awareness and cloud best practices in mind.
“This is going to be code generation based on your specific company code base,” Priyanka Vergadia, Google Cloud’s staff developer relations engineer, explained.
“This is not just code generation, but it also pulls up very specific metrics and functions from our product libraries," Vergadia said.
When reviewing the code generated by Duet AI, developers can also use a sort of find and replace function to make adjustments as needed without having to individually change potentially thousands of references to specific pieces of code.
Thus, Google executives argued onstage during the keynote, a process that currently takes months can be radically simplified. And, Bardoliwalla said, the tool can help cut down on misconfiguration and other code errors.
Duet AI can also be used to assist with root cause analysis when bugs pop up or sift through mounds of data in BigQuery.
The tool remains in preview, with general availability slated for later this year.
No more meetings?
Google had one more trick up its sleeve with Duet AI, integrating the tool into its Workspace family of apps with general availability. That means Duet AI can be used to as basically a virtual assistant. It can join Google Meet sessions to take notes for you, catch you up on parts of a meeting you may have missed or pull data that a superior requests.
There’s also an “attend for me” feature that allows Duet AI to act as a user’s stand in during a meeting, delivering a presentation for them and taking meeting notes.
Does this mean we may never have to attend a meeting again? Possibly. But if everyone in the meeting is a robot? Does that mean they’ll run the world?
Maybe, but that’s not what Google’s going for.
“This particular bots’ not going to contribute meaningfully to the conversation,” Kristina Behr, VP of products for Google Workspace, said in a pre-briefing call before Next. “We're still relying on humans and we see this as really an assistive technology to helping people collaborate so they can focus on the discussion or they can kind of catch up on the discussion quickly.”
Want to learn how to maximize your cloud investment? Check out our Cloud Executive Summit in beautiful Sonoma, Calif., from Dec. 6-7.