The development
What happened
Ai2 and University of Pennsylvania collaborators presented LitPivot on 23 October. The AI-assisted research-proposal system generates critiques and suggested revisions linked to excerpts from researcher-selected literature. Researchers control which papers inform an assessment and can accept, edit or reject its proposed changes.
Key facts
LitPivot separately organizes evidence for a proposal’s problem, contribution and evaluation plan, using Asta Paper Finder.
Users can accept, edit or reject suggestions; the system reranks literature and flags passages affected by a changed idea.
In a 17-researcher comparison using the same papers, two blinded experts rated literature grounding higher than a text-editor/chat baseline.
Improved novelty and usefulness were participant perceptions, not the experts’ assessment criterion. Five additional researchers tried ongoing projects in 30-minute sessions.
The context
Why it matters
The workflow makes the choice of evidence part of revising an idea, rather than treating a generated critique as authoritative. The small studies support a narrow literature-grounding finding, not demonstrated scientific breakthroughs or general productivity gains.