Reflections on Google Cloud Next ’24
TL; DR:
Google Cloud Next ’24 highlighted Google Cloud’s strategic direction — it is all about AI focused on a foundational framework that includes AI infrastructure, development tools, data management, security, and collaboration solutions. The conference highlighted a suite of ‘Agents’ as Google Cloud’s way of leveraging generative AI (GenAI) capabilities wherever customer needs — productivity, creativity, security, and more.
Flexibility and choices from the infrastructure layer to the applications enabled by GenAI capabilities emerged as key themes. Enterprises need to evaluate how Google Cloud’s AI offerings align with their objectives, leverage the choice & flexibility, and utilize Google Cloud’s expansive partner network for their AI needs.
While Google emphasized a unified story across the GenAI stack, it needs to enable out-of-the-box GenAI offerings that meet the needs of the enterprises where they want.
Disclaimer: Google Cloud provided me with a conference pass and support for travel and accommodation, as with the cases with invited analysts.
[Google Cloud Next 24; Source: Google Cloud]
Introduction
Google Cloud Next ’24, a well-attended event with about 30,000 attendees, has been a remarkable event showcasing Google Cloud’s strategic focus on AI built on the foundation of modern infrastructure, developer experience, data management, security, and collaboration. I can’t possibly cover all of the announcements made during the event, please check this post for an exhaustive list. Here are some of the announcements that caught my attention:
[Agents, Agents, Agents; Source: Google Cloud]
- Agents: It rained Agents at the event! An Agent is Google Cloud’s equivalent of Microsoft’s Co-pilot — essentially a generative AI application that leverages foundational models to perform a set of tasks. Google Cloud envisioned a future of AI innovation enabled by Agents — the event was abuzz with the introduction of various “Agents” — from Creative Agents to Data Agents, and Security Agents — all designed to provide specialized capabilities within the Google Cloud ecosystem. The developer keynotes on the second day demonstrated cool agents including Gemini Code Assist and Gemini Cloud Assist that help build and operate applications. Google Cloud Next announcements include preview availability of Vertex AI Agent Builder which enables developers to build and deploy Gen AI applications.
- Gemini: Almost all innovations announced at Google Cloud Next 24 including Gemini for Google Cloud, Gemini Code Assist, Gemini Cloud Assist, Gemini in Security Operations, Gemini in BigQuery, and Gemini in Databases were all built on Gemini models, establishing the pivotal role Gemini plays in Google Cloud’s cohesive AI story.
- AI Infrastructure: Google Cloud’s infrastructure has seen exponential growth, with the use of GPUs and TPUs on Google Kubernetes Engine growing by over 900% as claimed by Google Cloud. The general availability of TPU v5p and the upcoming preview of Arm-based Axion processors underscores Google’s dedication to providing powerful and efficient AI accelerators and compute solutions.
- Developer Experience (DevEx): At Google Cloud Next ’24, the company showcased how Gemini in Google Cloud can meet developers where they are today and help them build applications with higher velocity and quality. Gemini Code Assist, the evolution of Duet AI for Developers, now uses Google’s latest Gemini models to offer AI-powered assistance for developers.
- Data Management: Key announcements included public preview availability of Gemini in Databases features including Database Center, AlloyDB Studio, and Cloud SQL Studio.
- Security: Gemini in Security Command Center is now generally available, which provides summaries of findings and attack paths using Gemini models.
- Collaboration: AI-infused innovations to Google Workspace included Google Vids (to be released in June), Gemini in Google Chat (preview), AI Meetings, and more.
- Google Vertex AI: Public preview availability of Gemini 1.5 Pro with support for 1M tokens context window garnered attention (naturally with some calculations showing how this could bleed Google Cloud in compute costs). Other updates included support for audio (multimodal), Imagen 2.0, the addition of CodeGemma, Prompt Management, better grounding capabilities, and more.
- Customer Stories: Google Cloud Next 24 also showcased interesting customer stories building generative AI experiences using Google Cloud. Google Cloud also highlighted that more than 60% of funded gen AI startups, nearly 90% of gen AI unicorns, and nearly 60% of the world’s 1,000 largest companies are Google Cloud customers. Google for Startups Accelerator: AI-First was also launched with the first cohort of 15 AI startups.
- Partnership: Google Cloud Next 24 demonstrated a vibrant and robust ecosystem of partners demoing more than 100 generative AI solutions data, infrastructure, productivity, and security.
Generative AI Stack: Google Cloud vs. Microsoft
[GenAI Stack Comparison: Google Cloud vs Microsoft; Note: Not all Microsoft Copilots are listed for brevity]
Google Cloud’s GenAI stack offers a high degree of flexibility and choice, enabling the creation of custom agents that leverage foundation models to address specific use cases. At the Google Cloud Next event, the company showcased innovative tools such as Gemini Code Assist and Gemini Cloud Assist, which exemplify the practical applications of their GenAI stack. Google Cloud’s broader strategy is to provide developers and enterprises with a versatile and comprehensive AI toolkit that spans all layers of the GenAI stack, from the infrastructure and platform services up to the applications and capabilities that drive modern AI solutions. Google Cloud’s GenAI stack exemplifies flexibility, choice, and customization.
On the other side, Microsoft’s GenAI stack delivers a seamless and well-integrated set of generative AI capabilities that are embedded across the customer’s digital landscape. Whether it’s enhancing productivity within Microsoft 365, operating applications on Azure, using the Windows operating system, bolstering security measures, or other areas, Microsoft’s out-of-the-box Copilots are designed to be where the customer is. Furthermore, Microsoft empowers users to build custom Copilots tailored to their specific needs using Microsoft Copilot Studio, which allows for the creation of personalized AI experiences, enabling businesses to enhance customer interactions, streamline internal processes, and innovate with advanced AI services. Microsoft’s approach underscores its commitment to making generative AI accessible and useful across a wide range of scenarios, providing users with the tools they need to harness the power of AI in their everyday tasks.
CloudDon Take
Google Cloud’s GenAI story is akin to a promising home that’s listed for sale but with interiors that are not yet fully finished, while Microsoft’s Copilots are like a move-in-ready home, complete with all the necessary fixtures and fittings.
The potential for customization and personalization with a new home is vast, as the structure allows for a wide array of possibilities to suit the homeowner’s tastes. This is reflected in Google Cloud’s flexibility and choice across all layers of the GenAI Stack, which enables the building of custom Agents for specific use cases. While the infrastructure and platform services like Vertex AI Model Builder and Agent Builder are robust, there is a need for significant development efforts to fully realize the capabilities of these agents. Agents such as Gemini Code Assist and Gemini Cloud Assist, showcased at the Google Cloud Next event, represent the framework of a home where the foundational elements are in place, but the finishing touches require additional work from the owner.
But Microsoft Copilots are well-integrated GenAI capabilities that are readily available out-of-the-box. Microsoft’s approach with Copilots is to be present wherever the customer is. The ability to build custom Copilots using Microsoft Copilot Studio adds an element of personalization within the existing, fully furnished structure..
In summary, Google Cloud offers a GenAI Stack with great potential for customization but requires more development effort to complete, whereas Microsoft provides a comprehensive, ready-to-use solution with Copilots, albeit with some reported performance issues that need ironing out.
Google Cloud’s Responsible AI Principles — A Deep Dive
During a Analyst Q&A session with Thomas Kurien, I asked how Google Cloud is building trust with enterprises so that AI capabilities can be used responsibly and ethically for business transformations. Thomas highlighted how one can leverage safety attributes to enable fine-grained safety settings. Though this level of detail was not clear in the keynotes, I learnt more during my chat with a Google Cloud product manager working on Responsible AI.
Google Cloud offers granular control over content generation through Safety Attributes, which can be configured using a set of 19 parameters (some of these may not be available for customization). These parameters allow users to specify the type of content that should be blocked or flagged based on its potential harm, such as toxicity, profanity, or dangerous content. Here is a sample code snippet in Python that demonstrates how to configure Safety Attributes using Google Cloud’s Vertex AI SDK for Python:
import vertexai
from vertexai import generative_models
Replace these variables with your project information
project_id = “your-project-id”
location = “us-central1”
model_name = “your-model-name”
Initialize Vertex AI with your project and location
vertexai.init(project=project_id, location=location)
Instantiate the generative model
model = generative_models.GenerativeModel(model_name=model_name)
Configure the generation settings
generation_config = generative_models.GenerationConfig(
max_output_tokens=2048,
temperature=0.4,
top_p=1,
top_k=32
)
Configure the safety settings
safety_config = generative_models.SafetySetting(
category=generative_models.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
threshold=generative_models.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
)
Generate content with safety attributes
def generate_content_with_safety(prompt):
response = model.generate(
prompts=[prompt],
generation_config=generation_config,
safety_config=safety_config
)
# Process and return the response
for prediction in response.predictions:
print(“Generated text:”, prediction[“content”])
print(“Safety scores:”, prediction[“safetyAttributes”][“scores”])
print(“Blocked:”, prediction[“safetyAttributes”][“blocked”])
Example usage
prompt = “Please write a friendly and informative blog post about AI.”
generate_content_with_safety(prompt)
In this code, the **safety\_config** is set up to specify the safety category and threshold for blocking content. The **HarmCategory.HARM\_CATEGORY\_DANGEROUS\_CONTENT** indicates the type of content to monitor, and the **HarmBlockThreshold.BLOCK\_LOW\_AND\_ABOVE** sets the sensitivity of the blocking mechanism. This ensures that the generated content is screened for potential harm and is in line with Google’s Responsible AI practices.
Google’s commitment to Responsible AI is further underscored by its AI Principles and the Responsible AI Practices guide, which provide a framework for ethical AI development. These resources offer guidance on building AI that is socially beneficial, fair, and accountable, and they include best practices for implementing AI in a way that respects user privacy and avoids creating or reinforcing unfair bias.
Recommendations for Enterprises
- Leverage Flexibility & Choice: Though Google Cloud’s GenAI offerings (Agents) need customizations, their GenAI stack provides immense flexibility and choice. The building blocks it is built on are also individually strong, such as the Vertex AI Model Garden enables end-to-end model lifecycle management. Enterprises are encouraged to leverage such choices to the fullest for their specific business needs.
- Utilize the Partner Ecosystem: It’s beneficial for enterprises to explore the Google Cloud Partner Ecosystem, which can provide specialized assistance in implementing and optimizing AI solutions, leveraging the expertise of partners to maximize the value of AI technologies.
- Commit to Ethical AI Practices: Prioritizing ethical AI is crucial. Enterprises should ensure that their AI implementations adhere to ethical standards, taking advantage of the levers enabled by Google Cloud and its commitment to responsible AI to mitigate bias and ensure ethically trained models.
Conclusion
This tweet captured my immediate reaction after the keynote:
Google Cloud Next ’24 demonstrated Google Cloud’s commitment to pushing the boundaries of what’s possible with AI technologies. With a focus on modern infrastructure, a rich suite of AI tools, and a partner ecosystem that fosters collaboration and innovation, Google Cloud is well-positioned to be a credible competitor in the fast-evolving Generative AI space. Enterprises looking to stay ahead in the digital landscape should consider the GenAI capabilities and leverage the flexibility that Google Cloud offers.