AI Product Development Fundamentals
Basic knowledge of the fundamentals when builing AI products as a domain expert.
Questions in this deck
Your AI model works perfectly on your test data but fails badly when real users try it. What is this problem called?
You're building an AI system to detect fraud in financial transactions. What approach should you take first?
When building AI systems, what does "feature engineering" primarily involve?
When building AI products, what is "bias" in the context of model performance?
What is the primary purpose of splitting your data into training, validation, and test sets?
What is the main advantage of using pre-trained models (like GPT or BERT) rather than training from scratch?
What is typically the most time-consuming phase when building AI products in practice?
You have a dataset with 1000 examples of normal behavior and 10 examples of the rare event you want to detect. What is this problem called?
What does it mean to have a "minimum viable product" (MVP) mindset when building AI systems?