Prompt-story mismatches
#3
by jam963 - opened
Hi,
While running an evaluation on this dataset, I have found numerous samples where the prompt does not match at least one of the written stories. Some examples:
| Comment ID | Supplied prompt | Story content |
|---|---|---|
k3u4xyf |
A mimic disguised as a tavern clock | A narrator and their parents in Fallout’s New Vegas |
k3u8uz9 |
Same mimic prompt | A narrator and their mother in the Pokémon world |
ll2v72t |
Changes to single people’s sexuality/gender | A professor introducing a class on laws and taxes |
jantltr |
A missing heir after a train derails in Monster Town | A marriage proposal complicated by a silver allergy |
I've verified the first two as mismatched against the original Reddit post, but have not done so for the remaining ~100 flagged samples I found.
Could you check whether this release preserves the intended prompt–story associations and whether the parent-post IDs are
reliable? In particular, are these artifacts of text recovery, or could there be an alignment issue during dataset
construction?
Thanks!