Difference between revisions of "Science Agents"
KevinYager (talk | contribs) (→Chemistry) |
KevinYager (talk | contribs) (→Commercial) |
||
| (9 intermediate revisions by the same user not shown) | |||
| Line 9: | Line 9: | ||
* 2019-07: [https://doi.org/10.1038/s41586-019-1335-8 Unsupervised word embeddings capture latent knowledge from materials science literature] | * 2019-07: [https://doi.org/10.1038/s41586-019-1335-8 Unsupervised word embeddings capture latent knowledge from materials science literature] | ||
* 2024-11: [https://doi.org/10.1038/s41562-024-02046-9 Large language models surpass human experts in predicting neuroscience results] | * 2024-11: [https://doi.org/10.1038/s41562-024-02046-9 Large language models surpass human experts in predicting neuroscience results] | ||
| + | |||
| + | ==Explanation== | ||
| + | * [https://tiger-ai-lab.github.io/TheoremExplainAgent/ TheoremExplainAgent: Towards Multimodal Explanations for LLM Theorem Understanding] ([https://arxiv.org/abs/2502.19400 preprint]) | ||
==Autonomous Ideation== | ==Autonomous Ideation== | ||
| Line 15: | Line 18: | ||
* 2024-12: [https://arxiv.org/abs/2412.14141 LLMs can realize combinatorial creativity: generating creative ideas via LLMs for scientific research] | * 2024-12: [https://arxiv.org/abs/2412.14141 LLMs can realize combinatorial creativity: generating creative ideas via LLMs for scientific research] | ||
* 2024-12: [https://arxiv.org/abs/2412.17596 LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context] | * 2024-12: [https://arxiv.org/abs/2412.17596 LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context] | ||
| + | * 2025-02: [https://arxiv.org/abs/2502.13025 Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networks] | ||
==Adapting LLMs to Science== | ==Adapting LLMs to Science== | ||
| Line 41: | Line 45: | ||
===Commercial=== | ===Commercial=== | ||
| + | * [https://sakana.ai/ai-scientist/ Sakana AI] | ||
* [https://www.cusp.ai/ Cusp AI]: Materials/AI | * [https://www.cusp.ai/ Cusp AI]: Materials/AI | ||
| + | * [https://www.lila.ai/ Lila AI]: Life sciences | ||
| + | * [https://www.autoscience.ai/ Autoscience] ([https://www.autoscience.ai/blog/meet-carl-the-first-ai-system-to-produce-academically-peer-reviewed-research Carl]) | ||
==AI/ML Methods in Science== | ==AI/ML Methods in Science== | ||
===Chemistry=== | ===Chemistry=== | ||
| + | * 2025-01: [https://www.nature.com/articles/s41578-025-00772-8 Large language models for reticular chemistry] | ||
* 2025-02: [https://www.nature.com/articles/s42256-025-00982-3 Image-based generation for molecule design with SketchMol] | * 2025-02: [https://www.nature.com/articles/s42256-025-00982-3 Image-based generation for molecule design with SketchMol] | ||
| + | * 2025-02: [https://www.nature.com/articles/s42256-025-00994-z Large language models for scientific discovery in molecular property prediction] | ||
===Biology=== | ===Biology=== | ||
| Line 58: | Line 67: | ||
* 2025-02: [https://arcinstitute.org/manuscripts/Evo2 Genome modeling and design across all domains of life with Evo 2] | * 2025-02: [https://arcinstitute.org/manuscripts/Evo2 Genome modeling and design across all domains of life with Evo 2] | ||
* 2025-02: [https://www.microsoft.com/en-us/research/blog/exploring-the-structural-changes-driving-protein-function-with-bioemu-1/ Exploring the structural changes driving protein function with BioEmu-1] | * 2025-02: [https://www.microsoft.com/en-us/research/blog/exploring-the-structural-changes-driving-protein-function-with-bioemu-1/ Exploring the structural changes driving protein function with BioEmu-1] | ||
| + | * 2025-02: [https://arxiv.org/pdf/2502.18449 Protein Large Language Models: A Comprehensive Survey] | ||
| + | |||
| + | ===Successes=== | ||
| + | * 2025-02: [https://arxiv.org/abs/2502.11270 Site-Decorated Model for Unconventional Frustrated Magnets: Ultranarrow Phase Crossover and Spin Reversal Transition] | ||
==AI/ML Methods co-opted for Science== | ==AI/ML Methods co-opted for Science== | ||
| Line 80: | Line 93: | ||
* 2025-01: [https://agi.safe.ai/ Humanity's Last Exam] | * 2025-01: [https://agi.safe.ai/ Humanity's Last Exam] | ||
* [https://github.com/OSU-NLP-Group/ScienceAgentBench ScienceAgentBench] | * [https://github.com/OSU-NLP-Group/ScienceAgentBench ScienceAgentBench] | ||
| + | * 2025-03: [https://huggingface.co/datasets/futurehouse/BixBench BixBench]: Novel hypotheses (accept/reject) | ||
=Science Agents= | =Science Agents= | ||
Latest revision as of 12:38, 12 March 2025
Contents
- 1 AI Use-cases for Science
- 2 Science Benchmarks
- 3 Science Agents
- 4 AI Science Systems
- 5 Impact of AI in Science
- 6 Related Tools
- 7 See Also
AI Use-cases for Science
Literature
LLM extract data from papers
AI finding links in literature
- 2019-07: Unsupervised word embeddings capture latent knowledge from materials science literature
- 2024-11: Large language models surpass human experts in predicting neuroscience results
Explanation
Autonomous Ideation
- 2024-09: Mining Causality: AI-Assisted Search for Instrumental Variables
- 2024-12: Thinking Fast and Laterally: Multi-Agentic Approach for Reasoning about Uncertain Emerging Events
- 2024-12: LLMs can realize combinatorial creativity: generating creative ideas via LLMs for scientific research
- 2024-12: LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context
- 2025-02: Agentic Deep Graph Reasoning Yields Self-Organizing Knowledge Networks
Adapting LLMs to Science
- 2023-06: Domain-specific chatbots for science using embeddings
- 2024-10: Personalization of Large Language Models: A Survey
- 2024-11: Adapting While Learning: Grounding LLMs for Scientific Problems with Intelligent Tool Usage Adaptation
AI/ML Methods tailored to Science
Regression (Data Fitting)
- 2024-06: Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data: training on (x,y) pairs enables inferring underlying function (define it in code, invert it, compose it)
- 2024-12: OmniPred: Language Models as Universal Regressors
Tabular Classification/Regression
Symbolic Regression
Literature Discovery
- FutureHouse
- Lumina
- Automated-AI-Web-Researcher-Ollama
- 2025-01: Search-o1: Agentic Search-Enhanced Large Reasoning Models (project, code)
Commercial
- Sakana AI
- Cusp AI: Materials/AI
- Lila AI: Life sciences
- Autoscience (Carl)
AI/ML Methods in Science
Chemistry
- 2025-01: Large language models for reticular chemistry
- 2025-02: Image-based generation for molecule design with SketchMol
- 2025-02: Large language models for scientific discovery in molecular property prediction
Biology
- 2018: AlphaFold
- 2021-07: AlphaFold 2
- 2024-05: AlphaFold 3
- 2023-03: Evolutionary-scale prediction of atomic-level protein structure with a language model (ESMFold)
- 2023-11: Illuminating protein space with a programmable generative model
- 2024-11: Sequence modeling and design from molecular to genome scale with Evo (Evo)
- 2025-01: Targeting protein–ligand neosurfaces with a generalizable deep learning tool (Chroma)
- 2025-01: Simulating 500 million years of evolution with a language model (ESM 3 model)
- 2025-02: Genome modeling and design across all domains of life with Evo 2
- 2025-02: Exploring the structural changes driving protein function with BioEmu-1
- 2025-02: Protein Large Language Models: A Comprehensive Survey
Successes
AI/ML Methods co-opted for Science
Mechanistic Interpretability
Train large model on science data. Then apply mechanistic interpretability (e.g. sparse autoencoders, SAE) to the feature/activation space.
- Mechanistic interpretability for protein language models (visualizer, code, SAE)
- Markov Bio: Through a Glass Darkly: Mechanistic Interpretability as the Bridge to End-to-End Biology (quick description, background info on recent bio progress)
- 2023-01: Tracr: Compiled Transformers as a Laboratory for Interpretability (code)
- 2024-12: Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models
- 2024-12: InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders
- 2025-01: Insights on Galaxy Evolution from Interpretable Sparse Feature Networks
- 2025-02: From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models
Uncertainty
- 2024-10: entropix: Entropy Based Sampling and Parallel CoT Decoding
- 2024-10: Taming Overconfidence in LLMs: Reward Calibration in RLHF
Science Benchmarks
- 2024-07: SciCode: A Research Coding Benchmark Curated by Scientists (project)
- 2024-11: AidanBench: Evaluating Novel Idea Generation on Open-Ended Questions (code)
- 2024-12: LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context
- 2025-01: Humanity's Last Exam
- ScienceAgentBench
- 2025-03: BixBench: Novel hypotheses (accept/reject)
Science Agents
Reviews
- 2024-10: Empowering biomedical discovery with AI agents
- 2025-01: A review of large language models and autonomous agents in chemistry (github)
Specific
- 2024-01-13: ORGANA: A Robotic Assistant for Automated Chemistry Experimentation and Characterization (video)
- 2024-06-19: LLMatDesign: Autonomous Materials Discovery with Large Language Models
- 2024-08-12: Sakana AI: AI Scientist; The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery (code)
- 2024-09-09: SciAgents: Automating scientific discovery through multi-agent intelligent graph reasoning (code)
- 2024-09-11: PaperQA2: Language Models Achieve Superhuman Synthesis of Scientific Knowledge (𝕏 post, code)
- 2024-10-17: Rapid and Automated Alloy Design with Graph Neural Network-Powered LLM-Driven Multi-Agent Systems
- 2024-10-28: Large Language Model-Guided Prediction Toward Quantum Materials Synthesis
- 2024-12-06: The Virtual Lab: AI Agents Design New SARS-CoV-2 Nanobodies with Experimental Validation (writeup: Virtual lab powered by ‘AI scientists’ super-charges biomedical research: Could human–AI collaborations be the future of interdisciplinary studies?)
- 2024-12-30: Aviary: training language agents on challenging scientific tasks
- See also: AI Agents > Deep Research
Science Multi-Agent Setups
AI Science Systems
- 2025-01: Dolphin: Closed-loop Open-ended Auto-research through Thinking, Practice, and Feedback
- 2025-02: Towards an AI co-scientist (Google blog post: Accelerating scientific breakthroughs with an AI co-scientist)
Inorganic Materials Discovery
- 2023-11: Scaling deep learning for materials discovery
- 2023-11: An autonomous laboratory for the accelerated synthesis of novel materials
- 2024-10: Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models (code, datasets, checkpoints, blogpost)
- 2025-01: A generative model for inorganic materials design
Chemistry
- 2023-12: Autonomous chemical research with large language models (Coscientist)
- 2024-11: An automatic end-to-end chemical synthesis development platform powered by large language models
Impact of AI in Science
- 2024-11: Artificial Intelligence, Scientific Discovery, and Product Innovation
- 2025-02: Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation
Related Tools
Literature Search
Data Visualization
- 2024-10: Data Formulator: Create Rich Visualization with AI iteratively (video, code)
- Julius AI: Analyze your data with computational AI
See Also
- AI agents
- Nanobot.chat: Intelligent AI for the labnetwork @ mtl.mit.edu forum
