2015486 - Improve memories generation sensitivity filter

Status: NEW

(Core :: Machine Learning: General, enhancement)

Description

[Chris DiPersio]

Evaluation shows approximately 10% of generated memories from all models align with sensitive topics and should have been filtered out by the sensitivity filter. In actual counts, this is about 1-2 memories per run depending on the history size. We need to improve the performance of the filter to reduce this count.

This task likely overlaps with other sensitivity filtering efforts, like improving the topic block list and adding other regex-based filters.