AIncorrect
Hyperparameter tuning
Hyperparameter tuning is the work of adjusting training settings.
It is not the practice of designing the entire body of context information passed to a model, so this is incorrect.
An AI team is working on improving the quality of an agentic application. What is the name of the practice, increasingly emphasized in recent generative AI development, of designing the entire body of context information passed to a model (instructions, reference data, history, tool results, and so on) by deciding what to provide and in what order?
1 / 1
Select an answer
CorrectB
Explanation
A question about choosing the recent concept of designing the entire body of context information.
- 1「the entire body of context information passed to a model」Instructions, reference data, history, and so on = information provided as input
- 2「deciding what to provide and in what order」Designing the whole context = context engineering
BCorrect
Context engineering
Correct. Context engineering is the practice of designing the entire body of context information passed to a model (instructions, reference data, conversation history, tool results, and so on) by deciding what to provide, in what order, and in what amount. It has been emphasized recently as a broader concept than the prompt alone.
CIncorrect
Prompt engineering
Prompt engineering is the practice of refining how an individual prompt (instruction) is written.
The broader concept that refers to designing the entire body of context information, including instructions, reference data, history, and tool results, is context engineering, so this is incorrect.
DIncorrect
RAG (retrieval-augmented generation)
RAG is a method that retrieves external knowledge and uses it as grounding for answers.
It can be one of the materials provided to the context, but it is not the name of the practice of designing the entire context and in what order to provide it, so this is incorrect.
Key Takeaway
Note the correct answer, context engineering.
- The practice of designing the entire body of context information passed to a model (instructions, reference data, conversation history, tool results, and so on) by deciding what to provide, in what order, and in what amount.
- It is a broader concept than prompt engineering, which refines the wording of a single prompt, and has been emphasized recently.
Hyperparameter tuning (training settings), data labeling (adding labels), and quantization (lightening) are all not the design of context information.