The Google-led demonstration of Gemini Gen AI highlighted the profound impact AI will have on Workload Automation.

For the best moments, go to 2:30 and 14:20.

Transitioning Generative AI to Production

• Utilize Existing Solutions: Before building your own Large Language Model (LLM), explore available options like translation and transcription services. Don’t reinvent the wheel when robust solutions like Contact Center AI already exist.

Building LLM Applications

• Prompt Engineering: Start by crafting effective prompts. For LLMs to work on new cases, they need to be given the right context, often using Retrieval-Augmented Generation (RAG) to gather information.
• Long-Context Capabilities: With features like those in Gemini 1.5 Pro, you can input extensive information directly into prompts, simplifying the process.

Revolutionizing Development

• AI as an Assistant: In a fast-paced environment, AI can streamline the development process by organizing vast amounts of documentation and specifications into actionable insights.

Enhancing LLM Performance

To improve LLM performance, consider these strategies: clear instructions, assign a role, define goals and formats, include examples, add context, ask the model to explain its reasoning, break down complex tasks, and experiment with parameters.

LLM Grounding

• Grounding Techniques: When data exceeds the 2,000,000 token limit, grounding LLMs through RAG becomes vital. Using Vertex AI Search simplifies this process, allowing Broadcom teams to avoid building their own pipelines.
• Google Search Integration: Grounding LLMs with Google search results can reduce hallucinations and provide up-to-date information.

Understanding AI Agents

• AI Agents Defined: AI agents are autonomous applications with roles and tools designed to achieve specific goals. They consist of models (like Gemini 1.5 Pro), tools (APIs, functions), and orchestration to coordinate their efforts.
• Creating Agentic Applications: Utilize Vertex AI Agent Builder for a low-code solution to develop agent applications.

Multi-Agent Orchestration

• Teamwork in AI: Envision a team of agents, each with distinct roles. This collaboration allows for efficient task management and can outperform single advanced models in complex requests.

The Future of AI: Project Astra

• Innovative AI Assistants: Google’s Project Astra promises AI assistants that communicate using sounds, videos, and images. This includes recognizing sounds, answering queries, and even locating lost items.
• Personalized Experiences: The potential for tailored generative AI applications based on user interactions.
• The video of this is worth watching. Timestamp: 14:20

Partnership Highlights

• Google is a significant customer of Broadcom semiconductors.
• Broadcom is a significant Google customer.
• The companies share executive relationships for strategic alignment.
• Mutual technology use enhances product capabilities and service offerings.
• Broadcom’s products, like Automic SaaS, are listed on Google’s marketplace.
• Google provides services like GCVE (VMware), owned by Broadcom.