Difference between revisions of "Science Agents"

From GISAXS
Jump to: navigation, search
(→‎Genuine Discoveries)
(→‎AI Science Systems)
 
(9 intermediate revisions by the same user not shown)
Line 59: Line 59:
 
* 2025-08: [https://arxiv.org/abs/2508.15763 Intern-S1: A Scientific Multimodal Foundation Model]
 
* 2025-08: [https://arxiv.org/abs/2508.15763 Intern-S1: A Scientific Multimodal Foundation Model]
 
* 2025-11: [https://pubs.aip.org/aip/jcp/article/163/18/184110/3372267/A-foundation-model-for-atomistic-materials A foundation model for atomistic materials chemistry]
 
* 2025-11: [https://pubs.aip.org/aip/jcp/article/163/18/184110/3372267/A-foundation-model-for-atomistic-materials A foundation model for atomistic materials chemistry]
 +
* 2025-11: [https://arxiv.org/abs/2511.15684 Walrus: A Cross-Domain Foundation Model for Continuum Dynamics]
  
 
===Regression (Data Fitting)===
 
===Regression (Data Fitting)===
Line 98: Line 99:
  
 
===Materials===
 
===Materials===
−
* 2024-12: [https://www.nature.com/articles/s41467-024-54639-7 Crystal structure generation with autoregressive large language modeling
+
* 2024-12: [https://www.nature.com/articles/s41467-024-54639-7 Crystal structure generation with autoregressive large language modeling]
 
* 2025-03: [https://arxiv.org/abs/2503.03965 All-atom Diffusion Transformers: Unified generative modelling of molecules and materials]
 
* 2025-03: [https://arxiv.org/abs/2503.03965 All-atom Diffusion Transformers: Unified generative modelling of molecules and materials]
 +
* 2022-11: [https://arxiv.org/abs/2511.19730 Training-Free Active Learning Framework in Materials Science with Large Language Models]
  
 
===Chemistry===
 
===Chemistry===
Line 198: Line 200:
 
** 2025-07: [https://arxiv.org/abs/2507.00964 Benchmarking the Discovery Engine] ([https://www.leap-labs.com/blog/how-we-replicated-five-peer-reviewed-papers-in-five-hours blog])
 
** 2025-07: [https://arxiv.org/abs/2507.00964 Benchmarking the Discovery Engine] ([https://www.leap-labs.com/blog/how-we-replicated-five-peer-reviewed-papers-in-five-hours blog])
 
* 2025-07: [https://www.preprints.org/manuscript/202507.1951/v1 Autonomous Scientific Discovery Through Hierarchical AI Scientist Systems]
 
* 2025-07: [https://www.preprints.org/manuscript/202507.1951/v1 Autonomous Scientific Discovery Through Hierarchical AI Scientist Systems]
 +
* 2025-12: [https://arxiv.org/abs/2512.16969 Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows]
  
 
===Inorganic Materials Discovery===
 
===Inorganic Materials Discovery===
Line 220: Line 223:
 
===Bio===
 
===Bio===
 
* 2025-07: [https://arxiv.org/abs/2507.01485 BioMARS: A Multi-Agent Robotic System for Autonomous Biological Experiments]
 
* 2025-07: [https://arxiv.org/abs/2507.01485 BioMARS: A Multi-Agent Robotic System for Autonomous Biological Experiments]
 +
 +
===Physics===
 +
* 2025-12: [https://arxiv.org/abs/2512.19799 PhysMaster: Building an Autonomous AI Physicist for Theoretical and Computational Physics Research]
  
 
==LLMs Optimized for Science==
 
==LLMs Optimized for Science==
Line 255: Line 261:
 
=Genuine Discoveries=
 
=Genuine Discoveries=
 
* 2025-11: [https://cdn.openai.com/pdf/4a25f921-e4e0-479a-9b38-5367b47e8fd0/early-science-acceleration-experiments-with-gpt-5.pdf Early science acceleration experiments with GPT-5]
 
* 2025-11: [https://cdn.openai.com/pdf/4a25f921-e4e0-479a-9b38-5367b47e8fd0/early-science-acceleration-experiments-with-gpt-5.pdf Early science acceleration experiments with GPT-5]
 +
* 2025-12 [https://andymasley.substack.com/p/ai-can-obviously-create-new-knowledge AI can obviously create new knowledge - But maybe not new concepts]
  
 
* '''Math:'''
 
* '''Math:'''
Line 260: Line 267:
 
** 2025-11: [https://arxiv.org/abs/2511.02864 Mathematical exploration and discovery at scale]
 
** 2025-11: [https://arxiv.org/abs/2511.02864 Mathematical exploration and discovery at scale]
 
** 2025-11: [https://www.nature.com/articles/s41586-025-09833-y Olympiad-level formal mathematical reasoning with reinforcement learning]
 
** 2025-11: [https://www.nature.com/articles/s41586-025-09833-y Olympiad-level formal mathematical reasoning with reinforcement learning]
 +
** 2025-12: [https://arxiv.org/abs/2512.14575 Extremal descendant integrals on moduli spaces of curves: An inequality discovered and proved in collaboration with AI]
 
* '''Physics assistance:'''
 
* '''Physics assistance:'''
 
** 2025-03: [https://arxiv.org/abs/2503.23758 Exact solution of the frustrated Potts model with next-nearest-neighbor interactions in one dimension via AI bootstrapping]
 
** 2025-03: [https://arxiv.org/abs/2503.23758 Exact solution of the frustrated Potts model with next-nearest-neighbor interactions in one dimension via AI bootstrapping]
 +
** 2025-12: [https://www.sciencedirect.com/science/article/pii/S0370269325008111 Relativistic covariance and nonlinear quantum mechanics: Tomonaga-Schwinger analysis]
 +
*** [https://x.com/hsu_steve/status/1996034522308026435?s=20 Steve Hsu], [https://drive.google.com/file/d/16sxJuwsHoi-fvTFbri9Bu8B9bqA6lr1H/view Theoretical Physics with Generative AI]
 
* '''Literature exploration:'''
 
* '''Literature exploration:'''
 
** 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery]
 
** 2025-11: [https://arxiv.org/abs/2511.02824 Kosmos: An AI Scientist for Autonomous Discovery]
Line 274: Line 284:
 
** 2023-07: [https://www.nature.com/articles/s41586-023-06415-8 De novo design of protein structure and function with RFdiffusion]
 
** 2023-07: [https://www.nature.com/articles/s41586-023-06415-8 De novo design of protein structure and function with RFdiffusion]
 
** 2025-11: [https://www.nature.com/articles/s41586-025-09721-5 Atomically accurate de novo design of antibodies with RFdiffusion]
 
** 2025-11: [https://www.nature.com/articles/s41586-025-09721-5 Atomically accurate de novo design of antibodies with RFdiffusion]
 +
** 2025-11: [https://deepmind.google/blog/alphafold-five-years-of-impact/ AlphaFold: Five years of impact]
 
* '''Material Discovery:'''
 
* '''Material Discovery:'''
 
** 2023-11: [https://doi.org/10.1038/s41586-023-06735-9 Scaling deep learning for materials discovery]
 
** 2023-11: [https://doi.org/10.1038/s41586-023-06735-9 Scaling deep learning for materials discovery]

Latest revision as of 14:24, 29 December 2025

AI Use-cases for Science

Literature

LLM extract data from papers

AI finding links in literature

(Pre) Generate Articles

Explanation

Autonomous Ideation

Adapting LLMs to Science

AI/LLM Control of Scientific Instruments/Facilities

AI/ML Methods tailored to Science

Science Foundation Models

Regression (Data Fitting)

Tabular Classification/Regression

Symbolic Regression

Literature Discovery

Commercial

Bio

AI/ML Methods in Science

Imaging

Materials

Chemistry

Biology

Medicine

See: AI_Agents#Medicine

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.

Uncertainty

Science Benchmarks

Science Agents

Reviews

Specific

Science Multi-Agent Setups

AI Science Systems

Inorganic Materials Discovery

Materials Characterization

Chemistry

Bio

Physics

LLMs Optimized for Science

Impact of AI in Science

Related Tools

Literature Search

Data Visualization

Generative

Chemistry

Science Datasets

Genuine Discoveries

See Also

Retrieved from "http://gisaxs.com/index.php?title=Science_Agents&oldid=8505"