Sign in

Advait @ COLM

@advaitdeshmukh.com
117 followers 574 following 10 posts

CS PhD at CU Boulder

PostsRepliesMedia
Advait @ COLM @advaitdeshmukh.com · 10/09/2026
3/7 Through open coding of real user edits, we developed a framework identifying what users were editing in stories (plot, character, setting, genre…) and how they edited them (add, remove, change, extend). We then applied it to map the narrative choices users made as they explored these trees.
A heatmap of edit targets X direction
120
Advait @ COLM @advaitdeshmukh.com · 10/09/2026
2/7 To study these forking paths in the wild, we built two datasets from WildChat: 📖 WildStories: 275,635 real user requests for stories. ✍️ WildEdits: 24,291 branching trees reconstructing how users revised those requests and explored alternatives, with 100,200 annotated prompt edit pairs.
A figure explaining the data processing pipeline
140
Advait @ COLM @advaitdeshmukh.com · 10/09/2026
1/7 In 1941, Borges imagined an impossible novel that follows all narrative branches at once. Today, we can observe chatbot users exploring narrative possibilities as they repeatedly edit story prompts. In our COLM 2026 paper, we studied this branching exploration via the WildChat dataset. 🧵
Paraphrased edit trees from the dataset
24312