AI Foundations and Concepts
What to know
Distinguish narrow AI (systems built for a defined task) from speculative general AI, and separate machine learning from rule-based automation. Within ML, know the three learning modes: supervised (labelled examples, e.g. classification and regression), unsupervised (pattern discovery such as clustering), and reinforcement (learning from reward signals). Generative AI - large language models in particular - is trained to predict the next token, which explains both its fluency and its tendency to produce confident but wrong output.
Key vocabulary: model, training data, inference, parameters, tokens, context window, hallucination, fine-tuning, retrieval-augmented generation (RAG), and foundation model. Understand the difference between fine-tuning (changing model behaviour via additional training) and RAG (supplying external documents at query time).
Traps
Questions often conflate training with inference, or claim a model "knows" facts rather than predicting plausible text. Another trap: treating a larger context window as a guarantee of accuracy, or assuming generative AI is deterministic - the same prompt can yield different outputs. Watch for options that describe AI as conscious or reasoning like a human.
Worked example
A support team wants answers grounded in its current policy manual. Fine-tuning would bake stale text into weights and require retraining on every update; RAG retrieves the relevant passages at query time, so the manual can be edited without retraining. Rule of thumb: changing knowledge → retrieval; changing style or task behaviour → fine-tuning.