AWS Certified AI Practitioner AIF-C01 Practice Question

In-context learning keeps the base foundation model unchanged and simply adds example inputs and desired outputs to the prompt at inference time. Because no parameters are trained, there is no need to provision GPUs, store checkpoints, or run a dedicated model endpoint-costs are limited to normal inference charges for the larger prompt. Pre-training from scratch and full fine-tuning both require substantial training compute and long-term storage of a custom model. Parameter-efficient fine-tuning (such as LoRA) lowers training cost compared with full fine-tuning but still incurs extra training and separate hosting expenses. Therefore, supplying examples in the prompt provides the lowest up-front cost.