mlsanityData to 14 Oct 2026

AI Can Learn Scientific Taste

arXiv 2603.14473 ↗First posted 15 Mar 2026cs.CLcited by 11

Not on the map. Why?

Abstract (arXiv)

Scientific discovery depends on expert judgement and foresight, which we call scientific taste: the ability to judge and propose research ideas with the potential for long-term scientific impact. Scientific taste is largely concentrated among highly experienced researchers, whose expertise is usually limited to a few specialised fields. If AI could learn scientific taste, it could reduce reliance on human experts and accelerate scientific discovery. Whether AI can learn this ability remains an open question. We introduce Reinforcement Learning from Community Feedback (RLCF) to learn judgement and ideation. Scientific Judge learns from community feedback, such as citations. Scientific Thinker learns to propose research ideas with high potential impact. Experiments show that Scientific Judge outperforms strong LLM baselines and that learned judgement generalises to future-year papers, other community metrics, and unseen fields. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than those proposed by baselines. These results suggest that AI can learn scientific taste, marking an important step towards AI systems that could help accelerate scientific discovery.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 48 of its 77 references that are in the map. First posted on arXiv 15 Mar 2026.

Builds on

48 of 77 references.

References · 48

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  3. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  4. 2020Learning to summarize from human feedbackcited by 2,588
  5. 2025Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learningcited by 1,444
  6. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  7. 2018Open3D: A Modern Library for 3D Data Processingcited by 721
  8. 2020Jukebox: A Generative Model for Musiccited by 542
  9. 2024Generative Verifiers: Reward Modeling as Next-Token Predictioncited by 480
  10. 2024MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineeringcited by 469
  11. 2024YOLOv11: An Overview of the Key Architectural Enhancementscited by 425
  12. 2025Agent Laboratory: Using LLM Agents as Research Assistantscited by 425
36 more
  1. 2024Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researcherscited by 363
  2. 2024SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuningcited by 355
  3. 2025DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environmentscited by 286
  4. 2025Inference-Time Scaling for Generalist Reward Modelingcited by 239
  5. 2023Can large language models provide useful feedback on research papers? A large-scale empirical analysiscited by 200
  6. 2025Unified Reward Model for Multimodal Understanding and Generationcited by 199
  7. 2025RM-R1: Reward Modeling as Reasoningcited by 154
  8. 2024Generative Reward Modelscited by 148
  9. 2024CycleResearcher: Improving Automated Research via Automated Reviewcited by 148
  10. 2024MARG: Multi-Agent Review Generation for Scientific Paperscited by 104
  11. 2025DeepReview: Improving LLM-based Paper Review with Human-like Deep Thinking Processcited by 100
  12. 2025The Invisible Leash: Why RLVR May or May Not Escape Its Origincited by 97
  13. 2024AgentReview: Exploring Peer Review Dynamics with LLM Agentscited by 93
  14. 2025Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learningcited by 92
  15. 2025Unified Multimodal Chain-of-Thought Reward Model through Reinforcement Fine-Tuningcited by 90
  16. 2021The Logic of Graph Neural Networkscited by 85
  17. 2025DeepScientist: Advancing Frontier-Pushing Scientific Findings Progressivelycited by 63
  18. 2025The Ideation-Execution Gap: Execution Outcomes of LLM-Generated versus Human Research Ideascited by 62
  19. 2025rStar2-Agent: Agentic Reasoning Technical Reportcited by 54
  20. 2025Can LLM feedback enhance review quality? A randomized study of 20K reviews at ICLR 2025cited by 47
  21. 2025Reward Reasoning Modelcited by 47
  22. 2025Game-RL: Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoningcited by 32
  23. 2013Quantifying Long-Term Scientific Impactcited by 29
  24. 2025DianJin-R1: Evaluating and Enhancing Financial Reasoning in Large Language Modelscited by 22
  25. 2025InnovatorBench: Evaluating Agents' Ability to Conduct Innovative LLM Researchcited by 19
  26. 2026ArenaRL: Scaling RL for Open-Ended Agents via Tournament-based Relative Rankingcited by 17
  27. 2024From Words to Worth: Newborn Article Impact Prediction with LLMcited by 16
  28. 2024Exploring the Compositional Deficiency of Large Language Models in Mathematical Reasoningcited by 15
  29. 2025aiXiv: A Next-Generation Open Access Ecosystem for Scientific Discovery Generated by AI Scientistscited by 9
  30. 2025STRUCTSENSE: A Task-Agnostic Agentic Framework for Structured Information Extraction with Human-In-The-Loop Evaluation and Benchmarkingcited by 8
  31. 2020Neural MOS Prediction for Synthesized Speech Using Multi-Task Learning With Spoofing Detection and Spoofing Type Classificationcited by 7
  32. 2024OpenCIL: Benchmarking Out-of-Distribution Detection in Class-Incremental Learningcited by 7
  33. 2025NAIPv2: Debiased Pairwise Learning for Efficient Paper Quality Estimationcited by 4
  34. 2025Formally Verified Neurosymbolic Trajectory Learning via Tensor-based Linear Temporal Logic on Finite Tracescited by 3
  35. 2026Can Deep Research Agents Retrieve and Organize? Evaluating the Synthesis Gap with Expert Taxonomiescited by 3
  36. 2025Self-Foveate: Enhancing Diversity and Difficulty of Synthesized Instructions from Unsupervised Text via Multi-Level Foveationcited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 11.

11 papers.

  1. 2026ARIS: Autonomous Research via Adversarial Multi-Agent Collaborationcited by 16
  2. 2026SoundnessBench: Can Your AI Scientist Really Tell Good Research Ideas from Bad Ones?cited by 12
  3. 2026GIANTS: Generative Insight Anticipation from Scientific Literaturecited by 8
  4. 2026ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference Outcomescited by 2
  5. 2026Deep Research of Deep Research: From Transformer to Agent, From AI to AI for Sciencecited by 1
  6. 2026The Emerging AI Paper-Review Arms Race: Adversarial Co-Evolution in Scholarly Publishingcited by 1
  7. 2026FAME: Forecasting Academic Impact via Continuous-Time Manifold Evolutioncited by 0
  8. 2026mimeo: Compiling Public Expert Corpora into Agent Skills and Testing What Transferscited by 0
  9. 2026Reinforcing Agentic Creativity in Scientific Ideation with Night Sciencecited by 0
  10. 2026TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Expertscited by 0
  11. 2026Learning Individual Scientific Taste with Privileged Hindsight Distillationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Contemporary AI lacks the imagination to diverge or negate in science
  2. 2026Debate as Reward: A Multi-Agent Reward System for Scientific Ideation via RL Post-Training
  3. 2026Learning to Ideate for Scientific Impact
  4. 2026EvoIdeator: Evolving Scientific Ideas through Checklist-Grounded Reinforcement Learning
  5. 2026Think-Probe-Respond: Improving Large Language Models as Judges of Research Idea Novelty
  6. 2024AI Knowledge and Reasoning: Emulating Expert Creativity in Scientific Research
  7. 2024The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
  8. 2026The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks