Google: Making AI Easier to Do

The Cloud Field Day 20 (#CFD20) group spent most of a day at Google,? listening to what Google Cloud has to offer in the Generative Artificial Intelligence (#GenAI) space.

Google staff presented a lot of superb information, encompassing a wide range of technical capabilities. With lots of good details. Far too much for me to summarize in one blog. I’m going to refer you to the session recordings, and to another delegate’s summary blog, concerning most of that.

This blog will mostly be about an important concept from Google, and my take on what this means for organizations and individuals (aka “What’s In It For Me”, WIIFM) who are trying to figure out how to build AI skills, what AI might do for them, and what some options are concerning working with Generative AI models.

Google Terminology

A mild dose of terminology definitions will be helpful here, as Google sometimes assumes you know what the following things are.

·?????? Gemini is a set of AI models developed by Google, of various sizes and capabilities for various use cases. https://deepmind.google/technologies/gemini/

·?????? Project Astra consists of exploratory AI assistants that work with multi-modal information. Interesting demos: https://deepmind.google/technologies/gemini/project-astra/

·?????? Vertex AI is a fully-managed unified AI development platform for building and using generative AI.

·?????? Google Cloud Run is a Google facility to quickly build apps or websites on a fully managed platform. Run fully managed front-end, back-end services, batch jobs, websites and apps, and queue processing workloads without managing infrastructure.

·?????? Google Titanium is Google’s ongoing project to get away from being tied to “the big fast server” (my wording), by offloading certain tasks to improve performance and cost.

Google’s AI Capabilities

Google has extended its various cloud resource offerings to add a plethora of AI-centric resources suitable to GenAI models.

Their underlying Titanium offering gets away from server-centricity with compute or I/O limitations. It does so by offering tiered offloads: smartNICs, hyperdisk block storage, Borg scheduling across distributed systems, etc. Benefits include far better CPU response for real-time workloads, faster networking and storage IOPS, for cost optimization and faster results.

In similar fashion, Google’s Kubernetes Environment (GKE) provides managed Kubernetes services for auto-scaling of containers and clusters. It supports running auto-scaling applications on Kubernetes without “deep Kubernetes” knowledge. This is key for AI since most large models rely on Kubernetes to scale up model size.

Google’s AI Hypercomputer leverages processors including GPU’s and Vertex and GKE to provide a fully-managed AI solution.

CFD20 delegate Jack Poller covered some of this in Is Google Cloud the Right Solution for Your Modern Infrastructure?

Guy Currier wrote about the AI aspect of that in https://techfieldday.com/2024/have-highly-performant-ai-your-way/

The key summary takeaway about all this (from my perspective) is:

Google Cloud offers a spectrum of compute/network/storage/security services, ranging from fully managed to fully self-managed, both for general computing and now for AI.

Several of the Google presentations went into the finer details of that.

Implications for AI Consumers

Google’s presentations were technology-first, focusing on what they offer and their highly impressive technical capabilities.

What does it all mean for organizations doing (or thinking of doing) AI?

Google provides a spectrum of capabilities, ranging from manual/self-managed to fully automated and Google-managed.

On the left of the above (cropped photo from the presentation) we have compute (etc.) raw resources. If you don’t want to deal with Kubernetes and scaling a model up, there are managed capabilities around that. There are API’s so you can consume resources, pick and choose. And for getting started, there’s fully managed AI services, allowing you to get experimenting in AI without having to deal with a lot of implementation details

For a summary of Google’s AI products, see https://cloud.google.com/products/ai .

Why Do We Care?

AI is hot technology right now. But AI is subject to various concerns, starting with cost, climate impact, hallucination, staff skills, and the question of hype versus what’s real.

In relation to that, Tech Field Day hosted a podcast discussion about Gen AI in general that resonated for me: https://gestaltit.com/podcast/stephen/genai-is-revolutionizing-the-enterprise/ . What exactly should a company or organization be doing concerning AI?

A lot of organizations and people are trying to figure out the implications of all the GenAI discussion for their market, their products, and how they do business. The big challenge is getting your arms around what might be possible, how likely the alternatives are to succeed, and cost/ROI.

In short: How much should you or your company invest in AI-related activity, and what should its focus be?

What’s exciting about Google’s presentations is the spectrum of possible AI engagement possible:

·?????? If you or your company wants to be all-in, but not have to invest in buildout and support for dedicated AI resources, wants to get active doing AI without waiting for buildout of AI capacity, or wants to quickly scale up on-prem capacity … then using Google Cloud could make a lot of sense.

·?????? Using Google Cloud for AI ?also might allow you to focus on AI tasks and research, and not on building and managing the infracture. I.e. staff for AI research, not for supporting AI infrastructure.

·?????? Note that Google has highly optimized its AI offerings for performance, including cost optimization, rapid access to storage and data, etc. That may mean that you can get better performance at lower cost by using Google than by doing it yourself.

·?????? Given the current high demand for AI GPU’s and related infrastructure components, and related costs, there is a time-to-research aspect to all this.

I suspect many organizations may want to become engaged with AI, but with lower cost and more probable ROI, trading risk for smaller wins and experience. That’s where I see particular value in using Google / an AI Cloud provider.

But for those looking for big risk/big payoff, Google might also be an answer: develop and test in-house or on Google, then use Google to scale up without incurring the costs of massive hardware scale-up.

As I was writing this, a somewhat related article has appeared: https://www.networkworld.com/article/2505474/ai-success-real-or-hallucination.html?utm_date=20240627135220

What Might be Possible?

That’s also where the next to last couple of Google’s presentations are highly relevant and impressive, and perhaps helpful from the “what could I do with this, what’s in it for me?” perspective – as compared to Google’s technical capabilities. Impressive demos, including how little coding is needed to load data into Google Cloud, run an AI model on it, and then interact with the model.

I’ve listed al the CFD20 Google presentation video recording links below, with *** marking the relevant sessions (Presenters Lisa Shen and Neama Dadkhahnikoo).

For a sample of how a few lines of code let you run an AI app on Google Cloud, see Ray Lucchesi’s blog: GCP Cloud Run and VertexAI .

Google has made it possible to explore GenAI applications in an extremely low-friction way.

Some early solid wins with GenAI have been in terms of human language interface (front-end), chatops, plus performing tasks such as finding or summarizing or formatting content. These may be time-saving and lower risk of “hallucinations” as well (or where problems are easily caught by human review).

Multi-media interaction is the capability that really catches my eye. I personally currently vastly prefer written materials rather than video/audio recordings, primarily because (a) I can read a LOT faster than video playback, and (b) I can easily search text. AI might help with this: being able to do a text query to find a part of a video could be much faster than skimming back and forth in the video. In the worst case it fails to find a match, or finds an incorrect match.

Thought example: reporter trying to write an article, needs to find parts of video where politician said certain things (approximately worded).

Other CFD20 Resources about Google Cloud AI

My fellow delegates have busily blogged (etc.) about the Google presentations. Here are the other Google items I know about at the time of this writing:

·?????? Jim Czuprynski: https://jimthewhyguy.com/2024/06/20/theres-no-genai-easy-button-but-google-cloud-helps/

·?????? Keith Townsend: Unlocking AI Potential: Insights Into Google Cloud’s Vertex AI Platform (video)

Show Me the Videos!

For your convenience, here is the CFD20 video list from Tech Field Day’s Google Cloud page . They are in reverse order from the order they were presented in: start viewing from the bottom up!

·?????? CFD20:?Cloud Field Day at Google Cloud Wrap-Up ?featuring?Bobby Allen

·?????? CFD20:?Google Kubernetes Engine – The Container Platform for AI at Scale from Google Cloud ?featuring?Brandon Royal

·?????? CFD20:?Google Cloud Run and GenAI Apps ?featuring?Lisa Shen ***

·?????? CFD20:?Google Cloud Vertex AI Platform ?featuring?Neama Dadkhahnikoo ***

·?????? CFD20:?Generate Storage Insights with Gemini in Google Cloud ?featuring?Manjul Sahay

·?????? CFD20:?Gemini Code Assist in Google Cloud ?featuring?Rakesh Dhoopar

·?????? CFD20:?Gemini Cloud Assist in Google Cloud ?featuring?Bobby Allen

·?????? CFD20:?Google Cloud Network Infrastructure for AI/ML ?featuring?Victor Moreno

·?????? CFD20:?AI Workloads and Hardware Accelerators – Introducing the Google Cloud AI Hypercomputer ?featuring?Ishan Sharma

·?????? CFD20:?Security in Google Cloud ?featuring?Glen Messenger

·?????? CFD20:?AI/ML Storage Workloads in Google Cloud ?featuring?Sean Derrington

·?????? CFD20:?Running Modern Workloads in Google Cloud ?featuring?William Denniss

·?????? CFD20:?Running Enterprise Workloads in Google Cloud ?featuring?Jeff Welsch

·?????? CFD20:?Google Cloud Overview and Cloud Field Day Introduction ?featuring?Bobby Allen

Alternatively, there’s a YouTube playlist in the normal order at https://www.youtube.com/playlist?list=PLinuRwpnsHacXRHGUDkI8JeiVLS57bOmN

Comments

Comments are welcome, both in agreement or constructive disagreement about the above. I enjoy hearing from readers and carrying on deeper discussion via comments. Thanks in advance!

Hashtags: #CloudFieldDay20 #CFD20 #GoogleCloud #Google #BlueAlly #CCIE1773 #PeterWelcher #CiscoChampion

FTC disclosure statement: https://www.blueally.com/disclosure-statement/

Twitter: @pjwelcher

LinkedIn: Peter Welcher

Mastodon:@[email protected]