Potentially impactful research: Unjournal AI-assisted prioritization dashboard (~prototype)
A new prototype automatically scores papers from NBER and arXiv for policy impact using AI models.
The Unjournal, a project focused on evaluating research, has released a public prototype of an AI-assisted dashboard designed to prioritize academic papers for review. Built by David Reinstein, the system automatically discovers recent papers from major sources including NBER, arXiv (economics), SSRN, Semantic Scholar, and even the EA Forum. It then uses AI models, primarily from the GPT-5.4 family, to score them against key criteria: decision relevance, prominence, timing value, and methodological potential. The domain is focused on economics, quantitative social science, and policy-relevant research.
The tool is explicitly labeled as preliminary, with the team noting that many current AI recommendations are 'mediocre' and not yet well-calibrated. As of mid-April 2026, the AI only analyzes paper metadata and abstracts, not full texts. The scores reflect the expected value of commissioning an independent review, not the papers' inherent quality. The long-term vision is a hybrid or 'centaur' model where AI and human prioritization provide mutual feedback, refining the selection process for the most impactful research. The team is sharing the prototype for transparency and actively soliciting public feedback to guide its development.
- Automatically discovers papers from 8+ sources including NBER, arXiv, and Anthropic Research
- Scores papers using GPT-5.4 family models based on 4 key prioritization criteria
- Aims for a hybrid 'centaur model' combining AI and human feedback for better decisions
Why It Matters
This could dramatically accelerate the discovery of high-impact, policy-relevant research by automating the initial triage for expert evaluation.