Answer-first summary for fast verification
Answer: Curate upstream data properly that includes manual review before it is fed into the RAG system
The question focuses on mitigating offensive text outputs in a RAG system. Option D (curating upstream data with manual review) is the most effective because it directly addresses the root cause by filtering out toxic content before it enters the system, preventing the model from generating inflammatory responses. This approach is supported by the community consensus (100% selected D) and detailed reasoning, such as eliminating toxicity through manual review and ensuring quality in sensitive contexts. Other options are less suitable: A (increasing update frequency) may introduce more toxic data faster, B (informing users) does not prevent the issue, and C (restricting access) does not inherently remove offensive content from the data sources.
Author: LeetQuiz Editorial Team
Ultimate access to all questions.
A Generative AI Engineer is working to improve RAG quality by addressing its inflammatory responses. Which action is most effective for mitigating the problem of offensive text generation?
A
Increase the frequency of upstream data updates
B
Inform the user of the expected RAG behavior
C
Restrict access to the data sources to a limited number of users
D
Curate upstream data properly that includes manual review before it is fed into the RAG system
No comments yet.