IPA IP 2025 Question 80
A question IPA Japan published in 令和7年度 ITパスポート試験 公開問題 問80, translated into English. ITPEC is based on the same exams.
In AI, which of the following is the term for something that has been trained on wide-ranging and large amounts of data, and has the characteristic that it can be adapted to various uses, such as text-generation AI, through fine-tuning or other such methods?
Explanation
Foundation model — a large AI model trained on broad, huge data and then adapted to many different uses.
A foundation model, such as a large language model (LLM), is pre-trained on vast and varied data, text, images or code from many sources, mostly by self-supervised learning. It is not built for one task: fine-tuning on a smaller dataset, or instructions in a prompt, adapt the same model to different uses, such as text-generation AI, translation, summarising or answering questions. One costly training run thus becomes the foundation of many applications.
Answer (c)
Why not others:
- •(a) — annotation is attaching correct labels to training data for supervised learning; it prepares data, it is not a model
- •(b) — an expert system answers questions in one field with rules written down from human experts’ knowledge; it is not trained on large data
- •(d) — a convolutional neural network is a deep learning architecture suited to images; it names a structure of network, not broad pre-training followed by adaptation to many uses
Key rule: A foundation model is pre-trained broadly once and then adapted, by fine-tuning or prompting, to specific uses; generative AI services are built on such models.
Basic Theory · AI-written, may contain errors
This question comes from the past questions IPA Japan publishes, translated into English by ITPEC Practice. ITPEC Practice is an independent study tool and is not affiliated with IPA or ITPEC. See the original paper (PDF) or Report an issue.