GCP Generative AI Leader Practice Question 61: Foundation Model Concept
Google Cloud Generative AI Leader · Test 7 · question 1 of 10
A business leader is briefing new hires on generative AI basics. She wants a simple definition of a "foundation model" that explains why the same model can power a chatbot, a code assistant, and an image captioning tool without being rebuilt each time.
Which statement best describes a foundation model?
Show answer and explanation
Answer:
A foundation model is trained once on a broad, diverse dataset at large scale, and that broad pretraining is what lets one model be adapted, with prompting, few-shot examples, or fine-tuning, to many different tasks instead of being purpose-built for just one. That is exactly why a model such as Gemini can draft text, explain code, and describe an image without separate models for each job.
- A describes a narrow, single-purpose model, the opposite of the broad adaptability that defines a foundation model. - C describes a traditional rules-based expert system, which has no learned representations and cannot generalize to new tasks. - D is wrong because foundation models are prized precisely because they do not need to be retrained from scratch for each new task; adaptation is far cheaper than retraining.
Reference: What are foundation models?