The overwhelmed person’s guide to Google Cloud: week of Nov 13
Head of Developer Media
A weekly curation of the most helpful blogs, exciting new features, and useful events coming out of Google Cloud.
The content in this blog post was originally published last week as a members-only email to the Google Cloud Innovators community. To get this content directly in your inbox (not to mention lots of other benefits), sign up to be an Innovator today.
New and shiny
Three new things to know this week
- You get a slice … and you get a slice! Google Cloud just demonstrated the world’s largest distributed training job for LLMs, using more than 50,000 TPU v5e chips to do it. Even better, it’s not just a demo; we also announced the general availability of Cloud TPU Multislice Training, so you too can shard out your training jobs like Oprah giving away cars.
- Stay classy. Autoclass is now available for existing Cloud Storage buckets, and to see how much money that could save you, I once again commend to you Champion Innovator Lukas Karlsson’s helpful Autoclass Calculator.
- Upgrade unafraid. Cloud SQL users can now upgrade to the faster, near-zero downtime Enterprise Plus edition in place, without doing a manual migration. The anticipated downtime is less than 60 seconds; I’ve been a DBA for applications that … had significantly longer connection timeouts than that.
The hot seat
Behind the scenes of Google Cloud with the people who build it
Why vector embeddings are like DNA
We just announced a bunch of improvements to Vertex AI’s Vector Search capability. Here to help us understand what that means and why it matters is the lead product manager for Vector Search, Eran Lewis.
Forrest: Google Cloud has actually had a purpose-built vector database for quite awhile (the product formerly known as Vertex Matching Engine). Can you tell us about why Google initially developed this product? How did it evolve out of our own infrastructure needs?
Eran Lewis: The tech underpinning Vector Search was developed by Google Research. It’s a key enabler of Google's core business, powering services such as search and recommendations for YouTube and the Google Play Store. We built Vector Search (formerly known as Matching Engine) to make our blazingly-fast and scalable vector similarity search solution available to all developers, as a cost-effective solution for any embedding-based apps.
Do you have an analogy you like to use to explain what vector embeddings are and how they are used in applications?
Imagine vector embeddings as the DNA of digital information. Just as DNA encodes the essence of a living organism into a unique sequence of genes, vector embeddings encode complex data like text, images, videos or sounds into concise, distinct numerical vectors. These embeddings capture the core characteristics of the data, allowing AI systems to recognize, compare, and retrieve different pieces of information with incredible speed and accuracy.
This process is akin to how biologists use DNA to identify and understand the relationships between different species, making embeddings a fundamental tool for tasks like personalizing user experiences or finding semantically similar texts in large datasets.
What does the rise of RAG (retrieval-augmented generation) mean for vector DBs?
The rise of the Retrieval-Augmented Generation (RAG) framework highlights the critical need for vector databases in augmenting Large Language Models (LLMs). Vector databases ensure that LLMs provide fast answers based on accurate, real-world information, especially crucial when dealing with large data volumes.
For RAG to work effectively, vector databases must be fast, deliver high relevance, and be able to scale. Vertex AI Vector Search stands out in this regard, offering fast, relevant data retrieval and the ability to handle vast datasets, making it a top choice for powering RAG-supported AI applications.
What are some of the features that most excite you about the new and improved Vertex AI Vector Search?
Vector Search is an ideal choice if you're looking for a cost-effective solution that's not only fast but also scales as your app load grows. It supports filtering, indexes your embeddings on-the-fly, and supports key privacy and security features. Getting started is a breeze, thanks to the new UI and documentation. Additionally, Vector Search seamlessly integrates with Vertex AI Embeddings, making it an excellent solution for indexing and retrieving multimodal and text embeddings.
The best way to understand Vector Search is to start playing with it, and you can do that in this handy quickstart.
Every week I round up some of my favorite links from builders around the Google Cloud-iverse. Want to see your blog or video in the next issue? Drop me a line!
- If you missed KubeCon NA in Chicago last week, Daniel Bryant has you covered with a typically thorough and informed set of takeaways.
- I always enjoy Pablo Arrojo’s explorations of the itchy edges of Cloud Spanner. Today he’s examining how deleted rows affect query latency.
- “A fresh Cloud Functions project is like baby hair,” says Nino Handler, before he teaches you what to do when your sweet baby grows up into a horrible, unkempt teenager in Cloud Functions Combed.
Learn and grow
Four ways to build your cloud muscles this week
- What’s next for text. BigQuery just announced a set of new text analysis and preprocessing capabilities; this blog takes you on a wild trip from regular expressions to ML inference to show what’s possible.
- In-console-able. Duet AI is in the Google Cloud Console now! Here are three ways it can save you time.
- SQL without equal. The PostgreSQL interface in Cloud Spanner sometimes gets slept on, but this blog will wake you right up.
- Lights, camera, pipelines. Would you like to see MLOps in action? Yes you would. You can do it on November 21. Bring popcorn.
One more thing
This is just silly and fun - I made a little crossword puzzle to help you test your knowledge of GKE node pools. Answers in next week’s issue!
Become an Innovator to stay up-to-date on the latest news, product updates, events, and learning opportunities with Google Cloud.