AIGP - Understanding how to govern AI deployment and use - Section 4.2

Understand and compare AI model types, including classic versus generative, proprietary versus open source, small versus large, and unimodal versus multimodal capabilities.

Compare AI model types across key dimensions: classic versus generative AI, proprietary versus open-source, small versus large models, and unimodal versus multimodal capabilities. Recognise how each characteristic affects suitability, governance overhead, and risk profile.

generative AIopen source AI modelslarge language modelsmultimodal AI

Practice question for this objective

Free sampleUnderstanding how to govern AI deployment and useeasy

During an AI literacy briefing, a colleague asks what people mean when they call a system a large language model. Which description best captures what a large language model is?

  • AA model with very many parameters, trained on large amounts of text, that predicts likely token sequences to generate language. Correct
  • BA program that stores a fixed dictionary of written phrases and returns the closest stored phrase to a query by exact keyword lookup.
  • CAny model that processes images and video at high resolution, no matter whether it works with written or spoken language at all.
  • DA rule-based engine in which engineers hand-code grammar rules so that every sentence the system writes follows author-written instructions.
Define a large language model as a high-parameter model trained on text that generates language by predicting token sequences. A large language model is characterised by scale and method. It has a very large number of parameters and is trained on extensive text, learning the statistical patterns of language. At inference it generates output by repeatedly predicting the most probable next token. This learned, probabilistic generation of language is what sets it apart from keyword lookup tables, image-only systems, and hand-coded rule engines.

Why A is correct: A large language model has a large parameter count learned from extensive text corpora and generates language by predicting probable token sequences, which is exactly what the term denotes.

Why B is wrong: Keyword lookup resembles older retrieval tools and is tempting, but a large language model generates text from learned patterns rather than returning stored phrases verbatim.

Why C is wrong: Handling images is tempting because large models are powerful, but a large language model is defined by working with language; pure image processing is a different capability.

Why D is wrong: Hand-coded grammar describes earlier symbolic systems and is a plausible confusion, but a large language model learns statistical patterns from data rather than executing authored rules.

See more AIGP practice questions, answers explained.

More in this domain

Back to all Understanding how to govern AI deployment and use objectives, or the AIGP cert hub.

Examworthy is not affiliated with or endorsed by IAPP. Original, blueprint-aligned practice material only.