Foundations of AI

Master the fundamentals of large language models and generative AI, including text-to-text generation, multimodal capabilities, and training with rag and LMS vector stores for enterprise use.

Explore practical uses of large language models across business applications, from customer support chatbots and feature summarizations to code generation and vibe coding for rapid, visual app development.

Compare first-party models like OpenAI and Google with open-source options such as Meta Llama and Mistral. For general tasks, most models suffice, while coding favors Sonnet 3.7 and Gemini 2.5.

Explore pricing models for llms across major vendors like Google, Anthropic, and OpenAI, including token-based costs for prompts and outputs, and text-to-image pricing, with guidance on optimizing spend.

Explore prompt engineering basics to craft clear, context-rich prompts that improve model performance and yield better results, backed by scientific benchmarks and combined with creative techniques.

Explore zero-shot prompting and few-shot prompting as prompt engineering techniques, showing how adding examples narrows the model's vocabulary and improves results in tasks like classifying movie reviews.

Explore chain of thought prompting to elicit step-by-step reasoning from models, explain your reasoning process, and improve results for tasks that require reasoning and for agents discussed later.

Craft explicit prompts with rich context and examples to guide large language models, and specify what to avoid, to iteratively refine prompt engineering for ai products.

Explain retrieval augmented generation (rag) to reduce llm hallucinations by grounding answers in retrieved context from external data sources, like transcripts or real time data, via a prompt to grounded answer flow.

Use retrieval augmented generation to ground answers with base knowledge and company-specific policy documents, cite sources, and reduce hallucinations in customer service and product management.

Explore agents within retrieval augmentation generation, placing an LLM in the loop to filter relevant documents, ground answers, evaluate hallucinations, and iterate with multiple models for robust results.

Learn when to combine OpenAI and anthropic LLMs to optimize cost and speed, using a smaller model to evaluate outputs and grounding decisions within a flow-engineered agent.

Explore the rack pipeline's indexing, retrieval, and generation, showing how chunking data into meaningful pieces and embeddings build a knowledge base for semantically guided retrieval using vector similarity.

Explore how ai agents empower language models to act using pre-written tools. See how a chatgpt-like example drafts emails and fetches real-time data with tools.

Explore how language models use tools to act autonomously, guided by react-based prompt engineering, planning, and a human-in-the-loop for tasks like drafting emails or researching stock prices.

Equip an LMS with external tools and use its reasoning to decide when to act, enabling automation like drafting emails with data; 2025 is forecast as the year of agents.

Index data by chunking it into vectors and storing them in a vector store. Retrieve relevant information by querying the store to find vectors closest to the question.

Understand challenges in implementing rag systems, including data chunking size and meaning, secure indexing, evaluating chunks and answers, keeping data fresh, reducing latency, and managing context limits.

Provide relevant context to LLM prompts, such as current weather or up-to-date presidents, to prevent hallucinations and improve accuracy. Grounding and attribution then build trust by citing sources and dates.

Explore how embeddings convert text into high-dimensional vectors, reveal semantic similarity in a vector space, and store them in vector databases for efficient distance-based ranking.

Learn chunking by splitting large text into semantically meaningful chunks, such as questions and answers on a frequently asked questions page, then embed them into a vector store for retrieval.

Learn how fine-tuning tailors a large language model to a specific knowledge base, using prompt engineering or a dataset of questions and answers, with pros, cons, and cost considerations.

Fine-tuning requires extensive data preparation, time, and computing resources, and deploying the model adds hosting costs. Changing data forces restarting the process, risking overfitting and degradation on other subjects.

Large language models trained on internet data reflect overrepresented topics, shaping outputs, while what is not on the internet may not be represented, introducing bias in history, science, and beyond.

Explore the ethics of large language models, including misinformation, intellectual property, consent, accountability, and safety guardrails that prevent misuse and guide responsible deployment.

Learn how the European AI Act requires clear disclosure that content is AI-generated and guides developers to assess risk and safety of LLM applications, with fines for non-compliance.

Explore agi, artificial general intelligence, where an autonomous agent can take a complex human task, break it into steps, and complete the entire process requiring reasoning and planning.

Compare the AI product development lifecycle with the traditional lifecycle, noting AI adds a layer with large language models, unstructured and non-idempotent outputs, and the need for mitigations.

Explore best practices for deploying AI-enabled products, including testing, risk assessment, and disclosure of AI involvement, with caution for high-stakes applications and openness about potential hallucinations.

Discover how metrics for evaluating AI systems differ from traditional software metrics and explore using language models as judges to assess AI outputs, including latency considerations.

Understand how data quality and preparation, including cleaning to avoid garbage in garbage out, impact AI performance, and how AI can generate synthetic and tabular datasets.

Discover how AI-assisted testing integrates into the development life cycle by letting AI write tests for new code, run them automatically after each code generation, and fix issues it creates.

Explore Google's artificial intelligence stack—from research and Gemini to Gemini Code Assist, TPUs, and Vertex—and learn how product managers shape artificial intelligence strategy, readiness, and winning opportunities.

Celebrate foundations of ai and your generative ai course, and keep learning by staying curious, applying what you learned, asking questions on the student dashboard.