Google Professional Machine Learning Engineer Question 205
Select 4Google Cloud PlatformYou are tasked with evaluating a newly developed generative AI model for creating product descriptions for an e-commerce platform. The model generates text based on input product specifications. Which criteria should you prioritize to ensure the solution meets business and technical requirements?
- A
Relevance of the generated descriptions to the input product specifications
- B
Fluency and grammatical correctness of the generated text
- C
The number of layers in the neural network used by the model
- D
Bias and fairness in the generated descriptions
- E
Latency of generating product descriptions at scale
- F
Accuracy of the model's predictions compared to a labeled dataset
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Evaluating a generative AI solution requires focusing on both the quality of the outputs and the operational performance. Relevance, fluency, and fairness ensure the generated content meets business goals and ethical standards. Latency is important for scalability and user experience. While the architecture of the model and traditional accuracy metrics may be important during development, they are not primary evaluation criteria for generative AI solutions in production.
- A. Correct.
Ensuring the generated descriptions are relevant to the input specifications is critical for business use cases, as irrelevant descriptions can confuse customers and harm the platform's reputation.
- B. Correct.
Fluency and grammatical correctness ensure that the output is professional and readable, which is essential for improving user trust and engagement.
- C. Incorrect.
The number of layers in the neural network is a model architecture detail and does not directly impact the evaluation of the model's outputs in the context of business requirements.
- D. Correct.
Bias and fairness are key considerations to ensure that the model generates descriptions that are inclusive and do not propagate harmful stereotypes, which could lead to reputational damage or compliance issues.
- E. Correct.
Latency is vital to ensure the solution performs efficiently at scale, especially for an e-commerce platform where real-time or near-real-time responses may be required.
- F. Incorrect.
Accuracy, in terms of predictions against a labeled dataset, is not directly applicable here since generative AI models do not operate on labeled datasets in the same way as traditional classification or regression models.