IPA IP 2025 Question 28

It generates new content from natural-language instructions, on the basis of data learned in advance — that is generative AI.

Generative AI, such as large language models for text and image generation models, is trained beforehand on very large amounts of text, images or audio and learns their patterns. Given an instruction (a prompt) in natural language, it produces text, images, audio or program code that did not exist before, following those patterns. Its output can contain plausible-looking errors and can echo its training data, so it has to be checked before use.

Answer (d)

Why not others:

• •(a) — no learning at all is the opposite of how it works: generative AI depends on prior training on large datasets
• •(b) — a model that learns correlations from past data and predicts future values from current data is predictive AI (forecasting, regression); it outputs an estimated value, not new content
• •(c) — software robots that follow created scenarios are RPA; they repeat defined steps and generate nothing

Key rule: Generative AI creates new content from a prompt using what it learned in training; predictive AI estimates values from data; RPA repeats scripted operations.

Basic Theory · AI-written, may contain errors