Research map / Large language models / Agents, code and math
SciPredict: Can LLMs Predict the Outcomes of Scientific Experiments in Natural Sciences?
Udari Madhushani Sehwag, Elaine Lau, Haniyeh Ehsani Oskouie, Shayan Shabihi, Erich Liang, Andrea Sarai Echeverria Toledo, Guillermo A. Mangialardi, Sergio Fonrouge, Ed-Yeremai Hernandez-Cardona, Paula Vergara, Utkarsh Tyagi, Chen Bo Calvin Zhang and 5 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| provide expert-curated background knowledge in context (ablation) | all 15 LLMs · SciPredict | accuracy: about +3% on average (1.2 to 5.8% by model) | Improved |
| let the model generate its own background knowledge before answering (ablation) | LLMs · SciPredict | accuracy | Hurt |
| add self-generated background to expert background (SABK) (ablation) | LLMs with expert BK · SciPredict | accuracy | No clear effect |
| keep facts the model already knows in the background context (full vs filtered BK) (ablation) | LLMs · SciPredict | accuracy | Improved |
| convert MCQ questions to free-form by removing options (ablation) | LLMs · SciPredict MCQ subset | accuracy | Hurt |
| background knowledge effect on model confidence (ablation) | LLMs · SciPredict | self-reported confidence | No clear effect |
| use model self-reported confidence/feasibility to select tasks | LLMs · SciPredict | accuracy vs self-reported level: about 20% regardless of rating | No clear effect |
| use human-expert-rated feasibility to select tasks | Human experts · SciPredict | accuracy: about 5% -> about 80% | Improved |
| change decoding temperature 0.0 to 1.0 and top-p (ablation) | LLM predictions · SciPredict | accuracy | No clear effect |
| switch LLM judge from Gemini-3-Pro to GPT-5.2 (ablation) | free-form evaluation · SciPredict free-form | accuracy | No clear effect |
Closest papers by meaning
- PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors (NeurIPS 2025 · similarity 0.87)
- SciCode: A Research Coding Benchmark Curated by Scientists (NeurIPS 2024 · similarity 0.87)
- Measuring Scientific Capabilities of Language Models with a Systems Biology Dry Lab (NeurIPS 2025 · similarity 0.86)
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.85)
- Predicting Empirical AI Research Outcomes with Language Models (NeurIPS 2025 · similarity 0.85)
- LLM and Simulation as Bilevel Optimizers: A New Paradigm to Advance Physical Scientific Discovery (ICML 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
