Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Fine-tuning language models on journal publication decisions enables them to judge untested scientific ideas better than frontier models and human experts.result 90%
• Eleven frontier language models achieve only 31% accuracy on judging research pitch quality, barely exceeding chance.result 90%
• Panels of journal editors and editorial board members achieve 42% accuracy on judging research pitch quality.result 90%
• The best fine-tuned model achieves 59% accuracy on judging research pitch quality, surpassing all frontier models and human expert panels.result 90%
• Fine-tuned models exhibit calibrated confidence, achieving 100% accuracy on their highest-confidence predictions.technical 90%
• The method generalizes across disciplines, with models trained on economics publication records achieving 70% accuracy.result 90%
• Scientific taste can be extracted from institutional records of publication decisions rather than being explicitly articulated.method 85%
• The approach provides a scalable mechanism to triage scientific production in domains where quality resists formal verification.market 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card