Synthia: Visually Interpreting and Synthesizing Feedback for Writing Revision
summary
Synthia breaks dozens of free‑text critiques into interactive feedback bubbles that cluster, color, and resize by issue or actionability, links each unit to source sentences with bidirectional highlighting, and lets users combine bubbles to auto‑generate alternative drafts (fine‑tuned GPT‑3.5) and compare versions. Automated classification reaches F1 = 0.84, expert-checked target detection is high, and human verification finds 84% resolved of revision intents. A within-subjects study (N=12) shows more identified helpful feedback, more versions explored, and higher ownership than a GPT-4 chatbot baseline.
AI-written summary · about
chapters (10)
- 0:00Introduction to Synthia
- 1:49Feedback challenges in writing
- 2:55Design goals for Synthia
- 3:59Visualizing feedback prioritization
- 4:42Mapping feedback to text
- 5:49AI prototyping for revisions
- 6:51Technical evaluation results
- 7:55User study findings
- 9:20Key takeaways and future work
- 10:01Q&A session
related talks
37mBuilding AI-powered user interfaces
45mWeb Development in the Age of AI
35mBuilding AI-powered user interfaces
47mThe Art of Prompt Engineering: Best Practices for Effective AI Interactions
01minteresting experience with ChatGPT
50mIdentifying and fixing Issues in Code using AI based tools
29mSupercharge Your Productivity: Mastering Prompt Engineering
01h 03mKeynote: Red, Blue, and Purple AI