Research map / Large language models / AI and society
DeepScientist: Advancing Frontier-Pushing Scientific Findings Progressively
Yixuan Weng, Minjun Zhu, Qiujie Xie, QiYao Sun, Zhen Lin, Sifan Liu, Yue Zhang
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 |
|---|---|---|---|
| add surrogate-model plus UCB idea selection before implementation (ablation) | DeepScientist · Idea success rate | success rate: about 0% -> approximately 1-3% | Improved |
| scale parallel GPUs with shared Findings Memory (ablation) | DeepScientist · Progress findings in one week | number of SOTA-surpassing findings: one (4 GPUs) -> eleven (16 GPUs) | Improved |
| autonomous redesign by DeepScientist (A2P) over All-at-Once | All at Once · Who&When | accuracy: +183.7% | Improved |
| autonomous redesign by DeepScientist (ACRA) over TokenRecycling | TokenRecycling · MBPP | tokens/second: 190.25 -> 193.90 | Improved |
| autonomous redesign by DeepScientist (T-Detect, TDT, PA-Detect) over FastDetectGPT | FastDetectGPT · RAID | AUROC: +7.9% | Improved |
| DeepScientist papers vs other AI Scientist systems under automated review | DeepScientist · DeepReviewer-14B | rating: 5.90 vs best baseline 4.63 | Improved |
| DeepScientist papers vs ICLR 2025 average under human review | DeepScientist · Human program committee | rating: 5.00 vs 5.08 | No clear effect |
| human review of individual generated papers | PA-Detect and ACRA · Human program committee | rating: 4.33 vs 5.67 for TDT and A2P | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CycleResearcher-12B — CycleResearcher: Improving Automated Research via Automated Review (ICLR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Scientific Discovery with Generative AI: Progress, Opportunities, and Challenges (AAAI 2025 · similarity 0.89)
- AI-Researcher: Autonomous Scientific Innovation (NeurIPS 2025 · similarity 0.88)
- SciResearchBench: Benchmarking AI Agents on Complex Scientific Literature Discovery (NeurIPS 2026 · similarity 0.87)
- AutoScientists: Self-Organizing Agent Teams for Long-Running Scientific Experimentation (NeurIPS 2026 · similarity 0.86)
- DiscoveryWorld: A Virtual Environment for Developing and Evaluating Automated Scientific Discovery Agents (NeurIPS 2024 · similarity 0.86)
- AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite (ICLR 2026 · similarity 0.85)
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.
