問80
頻出度★★☆
AI において, 広範囲かつ大量のデータで訓練されたものであり, ファインチューニングなどによって文章生成 AI のような様々な用途に適応できる特徴をもつものを何というか。
令和7年度 ITパスポート試験 問80
テクノロジ系基礎理論 › 情報に関する理論
This question's explanation and your answers so far.
解説
問 80
Incorrectア
Annotation refers to the work of attaching correct labels (tags or notes) to each element of the training data. It is “data preparation work” for training an AI, not the trained “model” itself that can be adapted to various uses, so it is wrong.
Incorrectイ
An expert system is a traditional type of AI that stores the knowledge of experts in a specific field as “if-then rules” and makes judgments and gives advice through inference. Its mechanism differs from a foundation model, which is characterized by pre-training on large amounts of data and general adaptation through fine-tuning, so it is wrong.
- ウ
正解 The correct answer is “ウ Foundation model.” A foundation model is a general-purpose AI model that is pre-trained on a wide range of large-scale data and can then be adapted, through fine-tuning and the like, to a variety of downstream tasks such as text-generation AI, image-generation AI, and question answering. The statements in the question, “trained on a wide range of large-scale data” and “adaptable to various uses through fine-tuning and the like,” are exactly the definition of a foundation model, so ウ is correct. Text-generation AI (a large language model) is also one example built on top of a foundation model and adapted to a specific use.
Incorrectエ
A convolutional neural network (CNN) is a type of specific neural network structure (architecture) used mainly for tasks such as image recognition. It does not represent the concept of a general-purpose foundation that can be broadly adapted to many uses; it is a method aimed at a specific task, so it is wrong.
ここがポイント
- Foundation model = a general-purpose AI model that is pre-trained on large amounts of wide-ranging data and can be adapted to a variety of downstream tasks through fine-tuning. Text-generation AI (a large language model) is its representative application.
- Distinguishing confusing terms: annotation (the work of labeling data), the expert system (rule-based AI), and CNN (a specific neural network structure for images) all differ from the essence of a foundation model as a “model that can be generally adapted.”
用語