Reddit AI safety study maps fragmented discourse

This is a strong capstone because it treats "AI safety" as discourse, not doctrine. The useful insight is not that the clustering worked, but that it separates several different publics that usually get flattened together.

• –The most negative themes are concrete and lived: job loss anxiety, synthetic content spam, school misuse, and trust collapse around specific labs.
• –Enterprise adoption and national progress read much more neutral or positive, which suggests Reddit's AI conversation is pragmatically mixed rather than uniformly alarmist.
• –Framing matters as much as topic: two clusters can share keywords and still imply very different interventions, so topic labels alone would miss the policy signal.
• –The repo is transparent enough to inspect and reuse, but the query-based Reddit sample still makes the map exploratory rather than representative.