Free Databricks-Generative-AI-Engineer-Associate Actual Exam Questions - Question 5 Discussion
outputs.
Which action would be most effective in mitigating the problem of offensive text outputs?
Maybe D makes the most sense since offensive outputs usually come from bad input data. Stopping the problem before it starts seems better than just warning users or limiting access.
D, because cleaning the data before use directly targets the offensive content source.
A/C? Limiting users (C) won’t stop offensive outputs if the data itself is toxic, and just updating data more often (A) doesn’t guarantee better quality. Tackling the root cause in data seems more effective.
D Manual review catches offensive content early, stopping it from ever influencing the model. It’s more direct and reliable than just updating data or limiting user access.
A vs D? While D focuses on manual review, which is ideal for catching problematic data before it’s used, it might not be scalable for very large datasets. A could help by ensuring the system uses the latest, less biased info, reducing outdated or offensive content naturally over time. So, if manual curation isn’t feasible due to volume, increasing update frequency might still make a noticeable difference in reducing inflammatory outputs.
A imo, updating data more often might help reduce outdated biases or errors that cause offensive outputs. Keeping the data fresh could prevent some inflammatory content without needing full manual review.
Option D because fixing data quality is the only real way to prevent bad outputs.
Makes sense that D is the best since cleaning the data upfront tackles the source of the offensive content directly. If the input data is full of inflammatory stuff, no amount of user warnings or limiting access will help. So focusing on quality control before feeding data into the system is the most solid fix here.
It’s D, but what’s the size and type of the upstream data?