Benchmark RadarAI BENCHMARK PROFILE
ProjectionBench
A framework for evaluating scientific hypothesis generation in LLMs under progressive information disclosure, comparing model hypotheses to original paper conclusions via semantic similarity.
- Released
- 2026-05-28
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
It assesses a model's innovativeness and grounded reasoning in scientific discovery, which is crucial for developing AI co-scientist systems.
Motivation
Scientific discovery is an inherently creative and uncertain process, requiring reasoning beyond the recall of known knowledge.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.