My understanding of how large language models work is rooted in my knowledge of statistics. However a significant number of people have never been to college and statistics is a required course only for some degree programs.

How should chatgpt etc be explained to the public at large to avoid the worst problems that are emerging from widespread use?

Roughly knowing how many times each tag has been used would provide users actionable information if they would like to search or filter by tags.

It might improve UX when applying tags, but might have undesirable side effects in user behavior.

I can think of three places this might be implemented, and I don't know which, if any, we want:

When filtering topics by tags:

• informs users how large or small their scope is
• this view should probably be kept somewhat up to date

When looking at a topic's tags:

• informs users where to start searching/filtering
• passively builds a frame of reference for how tags are used?
• this view could be allowed to become outdated and stale without issue

When applying tags

• a more common tag might be less accurate, but it might be more helpful?
• in the auto fill issue weight by frequency was proposed, which is somewhat similar but more opaque
• this should probably use pretty recent counts as well