Generative AI Leader Study Plan¶
Three weeks at 4-5 hours per week. Shorter than most plans here, because there is no lab work and the technical depth is deliberately limited.
Week 1: Fundamentals¶
- Download the exam guide and the official study guide
- Start the free Cloud Skills Boost learning path
- AI, machine learning, deep learning, and generative AI: how they nest
- Foundation models, and how they differ from task-specific models
- Tokens, context window, and how both drive cost
- Embeddings and semantic similarity
- Multimodal models: text, image, audio, video
- Pre-training, fine-tuning, and instruction tuning
- Hallucination, its causes, and grounding as the mitigation
- Agents and tool use at a conceptual level
- Model limitations: knowledge cutoff, reasoning limits, bias, non-determinism
- Review Notes:
notes/01-genai-fundamentals.md
Week 2: Google Cloud offerings and techniques¶
- Vertex AI: Model Garden, Vertex AI Studio, training, deployment, evaluation
- Gemini model family and multimodality
- Gemma open models; Imagen, Veo, Chirp
- Vertex AI Agent Builder and Conversational Agents
- Vertex AI Search for grounded enterprise search
- Gemini for Google Workspace and Gemini for Google Cloud
- NotebookLM
- BigQuery generative AI functions
- AI Hypercomputer, TPUs, GPUs
- Build versus buy: the ladder from applied assistant to custom model
- Prompt engineering: zero-shot, few-shot, chain-of-thought, system instructions
- Parameters: temperature, top-k, top-p, output length
- Grounding with Google Search and with enterprise data; RAG
- Fine-tuning, and when it is worth the cost
- Function calling and extensions
- Evaluation approaches and safety filters
- Review Notes:
notes/02-google-cloud-offerings.mdandnotes/03-improving-output.md
Week 3: Business strategy and review¶
- Identifying valuable use cases, and recognizing ones that do not need AI
- Prioritizing by value and feasibility
- Building the business case and defining success measures
- Total cost of ownership beyond inference
- Data readiness as the usual real constraint
- Change management, training, and adoption
- Responsible AI: fairness, transparency, explainability, privacy, safety, accountability
- Google's AI Principles and the Secure AI Framework
- Risks: hallucination in customer-facing use, IP, data leakage, regulation
- Human in the loop, and where automation is inappropriate
- Review Notes:
notes/04-business-strategy.md - Two timed practice exams
Readiness check¶
- Explain what a foundation model is and why it changed the economics of AI
- Explain grounding and how it reduces hallucination
- Choose between grounding, RAG, and fine-tuning for a stated requirement
- Name the Google product for enterprise search, conversational agents, and image generation
- Explain when to use an applied assistant rather than building on the platform
- Describe how you would measure whether a generative AI project succeeded
- List Google's responsible AI themes and give a control for each
- Name a use case that sounds like AI but should be solved another way