Princeton field study finds LLM templates raised feedback rates by 9.4 points
manoelribeiro · x · 2026-07-29
A Princeton talk previews a field experiment on AI-mediated feedback in higher education.
- In an undergraduate machine-learning course, teaching assistants were shown LLM-generated feedback templates after grading.
- TAs could edit or ignore the template; control TAs received nothing.
- Across 2,840 observations, AI mediation increased the share of students who received feedback by 9.4 percentage points and also increased feedback length.
- Student ratings of feedback quality stayed comparable across conditions, suggesting AI can raise discretionary effort without obvious quality loss.
The result is framed as a broader design pattern: using LLMs as mediators that augment expert judgment instead of replacing it.
More from Apps
- Codex plugs into Roblox Studio to build, test, and publish a game end to end — majidmanzarpour · 2026-07-29
- PenEcho adds AI reasoning, Mermaid flowcharts, and multi-model support to Excalidraw-style canvases — abhishek__AI · 2026-07-29
- A screenshot shows gpt-transcribe appearing in the OpenAI API — BLCNYY · 2026-07-29
- Viktor pitches a Slack-based AI employee that handles client decks, recaps and growth updates — Salmaaboukarr · 2026-07-29
- Invideo’s Agent One turns a cinematic trailer into a conversation-driven workflow — azed_ai · 2026-07-29
- Granola launches an Apple Watch app for hands-free meeting transcription — soleio · 2026-07-29