Free Practice Questions for IBM watsonx Generative AI Engineer v1 - Associate (C1000-185) Certification

• Define and describe the five key capabilities of generative AI and large language models (LLMs) o Summarization o Classification o Generation - Code - Translation o Extraction o Q&A • Review case studies or examples demonstrating each capability. • Evaluate the impact of these capabilities in various industries • Discuss potential future developments in LLM capabilities.

• Identify common patterns used in generative AI solutions. o Mixture of experts (MoE) o Variation Auto Encoders (VaE) o Transformer based models o Reasoning models • Describe the components that constitute these patterns (e.g., input data processing, model training, output generation). • Analyze real-world examples to illustrate how these patterns are applied. • Create a diagrammatic representation of various generative AI patterns.

• List the technical and ethical limitations of generative AI and LLMs. • Discuss scenarios where generative AI may produce biased or incorrect outputs. • Explore strategies to mitigate these limitations in practical applications. • Evaluate the risks associated with the deployment of generative AI in sensitive or critical applications.

• Study various industry sectors to identify potential use cases for generative AI. • Conduct needs analysis to determine how generative AI can address specific business problems. • Propose generative AI solutions for identified use cases. • Prepare feasibility reports for the proposed solutions.

• Analyze the requirements of a given use case. • Determine criteria for selecting an appropriate model based on performance, efficiency, cost, and ethical considerations. o Model selection based on Parameter size o Chat vs instruct o IBM Granite models o Billing classes • Simulate decision-making processes to select the optimal model for a use case.

• Define model architecture and its significance in AI solutions. • Examine various model architectures and their components. • Match model architectures with specific use cases based on technical requirements. • Agentic architectures

• Understand the RAG Pattern o Introduction to the RAG pattern to Basic Search o Demonstrate the RAG pattern with LangChain • AI agents

• Understand the Security Risks Associated with Inputs o Data Bias o Data Poisoning o Data Curation and Downstream Retraining o Data Privacy o Prompt Injection o Prompt Leaking • Understand the Security Risks Associated with Output o Output and Decision Bias o Revealing Confidential or Personal Information o Toxic Output o Spreading Miss disinformation and toxicity o Harmful code generation • Understand the Security and Privacy for Foundation Models • Guardian models

• Introduction to Zero-shot prompting • Introduction to few-shot prompting • Try Zero-shot prompt sample • Try Few-shot prompt sample

• Choose right model for a use case. • Try to converse with model for a chat use case. • Try to converse with model to translate from one language to another.

• Introduction to Prompt templates • Evaluate Prompt Templates • Creating Environment Templates • Deploying Prompt Templates • Tracking Prompt Templates

• Introduction to model parameters o Decoding o Greedy Decoding o Sampling Decoding • Understanding the Random Seed • Understanding the Repetition Penalty • Understanding the Stopping Criteria • Understanding the Stop Sequences • Understanding the Minimum and Maximum Tokens

• Articulate what a prompt variable is • Identify optimal static prompt text to replace with variable • Summarize the value of reusable prompts

• Articulate the prompt editing options o Chat o Structured o Freeform • Demonstrate how to build reusable prompts • Describe example input prompts and their use

• Articulate the decoding process at a high level o Greedy decoding o Sampling decoding • Describe the sampling decoding parameters • Temperature o Top K o Top P o Random Seed o Repetition penalty • Articulate stopping criteria and examples o Stop sequences o Minimum and Maximum tokens o model generation time limit

• Describe hallucinations in model output o Underlying causes o Techniques for avoiding • Articulate the risks associated with personal information o Identify Personal information o Techniques for excluding o PII filter • Hate speech, abuse, and profanity o Techniques for reducing risk o HAP filter • Bias o Underlying cause of bias in outputs o Techniques for reducing bias • Debate the risks of data bias and poisoning

• Differentiate between hard prompts and soft prompts o Designed by humans or AI o Readability of the prompt o Explainability • Articulate how soft prompts are generated • Debate the benefits and drawbacks of soft prompts o Performance o Simplicity o Interpretability o Explainability

• Manage the token usage of each prompt template and model • Detect inefficient prompt techniques • Design cost-effective prompt templates • Employ model parameters to reduce the generation cost o Stop sequences o Min/max token limits

• Add data to a watsonx.ai project for tuning • Inspect and validate data elements using Data Refinery

• Understand quantization is in the context of LLMs o Tradeoffs - Reduce precision - Reduce computational costs o Techniques

LoRA

• Summarize the purpose of taxonomy tree-based curation • Generate synthetic data using InstructLab o Describe LAB (Large-scale Alignment for chatBots) methodology

• Describe what the components of InstructLab o Taxonomy driven data curation o Large scale synthetic data generation o Iterative, large scale alignment tuning - Knowledge tuning - Skill tuning • Articulate the InstructLab workflow

• Describe the two options supported o Leverage your existing data o Create from your custom data schema • Understand limitations on importing existing data sources and size limitations. • Understand anonymization of imported data • Understand the two algorithms for mimicking existing data o Kolmogorov-Smirnov o Anderson-Darling • Understand differential privacy concepts and settings o Privacy budget o Privacy leakage probability o Random seed • Understand sizing requirements for synthetic data generator

• Understand concepts of text embeddings. • Describe different embedding models o IBM Embedding models o Third party embedding models

• Perform converting text to embedding vectors • Describe prerequisites for embedding API o Credentials o Project ID o Deployment ID • Understand the choices and capabilities of vector databases o Purpose built vector databases o Vector extensions to popular databases

• Understand the concept of a retriever • Describe different types of retrievers o Vector databases - Embedded - Static o Watson Discovery o GitHub code retrieval API o watsonx Discovery • Understand the capabilities of retrievers • Understand use case driven selection of a retriever.

• Understand the RAG Pattern • Understand implementation details of a RAG pattern o LangChain and WatsonX LLM o LangChain, Watson ML and ElasticSearch o LangChain, SingleStore o LlamaIndex • Understand chunking/text splitting • Agentic RAG • AutoRAG

• Define the deployment lifecycle of a prompt template • Manage changes to prompts templates in applications • Understand roles involved with deploying AI Assets • Understand need around AI Governance o Model performance o Evaluate inferences o Explain outcomes from prompts

• Benefits of deployments space o Govern lifecycle (dev/evaluation/production) o Separate endpoints for applications to call • Benefits when deploying a prompt template • Changes to application when using a deployment space

• Understand requirements of watsonx and foundation model • Application access to a custom model

• Version using deployment spaces • Changes to applications on a prompt version change • Testing new prompt template version • Model gateway

• Understand the options between managed Software as a Service or Software • Deployment choices based Architectural pattern o RAG o Summarization o Q&A • Corpus of Data repository o AI Pipelines to update Repository o Manage data in corpus o Adjusting text chunk size • Identify endpoint security and stability • Code generation within PromptLab

• Connect watsonx.ai to other IBM Cloud services, such as Watson Assistant for chatbot development or Watson Discovery for document understanding, to build end-to-end AI solutions. o Identify relevant IBM Cloud services that can be integrated with watsonx.ai (e.g., Watson Assistant, Watson Discovery). o Understand the integration points and data exchange mechanisms between watsonx.ai and the chosen services. o Configure the integration using watsonx.ai's UI or APIs. o Integrate with watsonx.governanance • Leverage watsonx.ai's APIs and SDKs to integrate generative AI models with external applications, platforms, or custom workflows, including utilizing LangChain for complex chain creation and management. o Obtain API keys or credentials for external applications or platforms. o Write code using watsonx.ai's SDKs to send requests and receive responses from the AI models. o Handle errors and exceptions during API interactions.

• Design and implement workflows utilizing watsonx.ai that combine multiple tasks, including LangChain-based chains, to orchestrate complex generative AI processes. o Break down a complex AI task into smaller, manageable steps. o Design a workflow using orchestration tools. • Utilize orchestration tools to automate the execution of workflows, including scheduling, dependency management, and error handling. o Schedule workflows to run automatically at specific times or intervals. o Implement error handling and retry mechanisms to ensure robustness. o Use conditional logic to trigger different workflow branches based on data or model results. • Monitor and troubleshoot workflow execution, identifying bottlenecks, failures, or performance issues. o View workflow logs and execution details within watsonx.ai. o Identify and resolve errors or bottlenecks that may occur during workflow execution. o Optimize workflow performance by adjusting parameters or parallelizing tasks. • Agentic RAG

• Given a business use case, design a comprehensive solution that integrates watsonx.ai with relevant data sources, other IBM Cloud services, and external systems. o Identify the necessary data sources, IBM Cloud services, and external systems to be integrated. o Design a high-level architecture diagram outlining the data flow and interactions between components. • Implement the designed solution, leveraging the appropriate watsonx components and tools. o Implement the integration using the chosen tools and technologies. o Write code to connect data sources, invoke watsonx.ai APIs, and interact with external systems. o Test the integration in a development environment. • Test and validate the integration, ensuring seamless data flow, reliable communication, and expected functionality. o Deploy the integrated solution to a production or staging environment. o Perform end-to-end testing to verify functionality and data flow. o Monitor the solution's performance and address any issues that arise. • Model Context Protocol

• Explain the core concepts of LangChain (chains, agents, tools, memory) and their role in building complex conversational AI and generative AI applications. o Define the core concepts of LangChain, such as chains, agents, tools, and memory. o Explain how LangChain simplifies the development of complex conversational AI and generative AI applications. o Describe the benefits of using LangChain, such as modularity, reusability, and extensibility. • Demonstrate the ability to create and customize LangChain chains, incorporating various components like LLMs, prompt templates, and external data sources. o Create a simple LangChain that combines an LLM with a prompt template. o Customize a LangChain by adding custom components, such as a tool for interacting with an external API. • Design and implement LangChain agents that can interact with the environment, make decisions, and take actions based on model outputs and external information. o Design a LangChain agent that can perform a specific task, such as answering questions or generating summaries.