Difference between revisions of "Science Agents"

From GISAXS
Jump to: navigation, search
(→‎AI Science Systems)
(→‎Mechanistic Interpretability)
 
(4 intermediate revisions by the same user not shown)
Line 11: Line 11:
 
* 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]
 +
 +
===(Pre) Generate Articles===
 +
* 2022-12: [https://aclanthology.org/2022.emnlp-main.296/ Re3: Generating Longer Stories With Recursive Reprompting and Revision]
 +
* 2023-03: English essays: [https://journal.unnes.ac.id/sju/index.php/elt/article/view/64069 Artificial intelligence (AI) technology in OpenAI ChatGPT application: A review of ChatGPT in writing English essay]
 +
* 2023-01: Journalism: [https://journals.sagepub.com/doi/10.1177/10776958221149577 Collaborating With ChatGPT: Considering the Implications of Generative Artificial Intelligence for Journalism and Media Education]
 +
* 2023-07: Science writing: [https://www.rbmojournal.com/article/S1472-6483(23)00219-5/fulltext Artificial intelligence in scientific writing: a friend or a foe?]
 +
* 2024-02: Wikipedia style: [https://arxiv.org/abs/2402.14207 Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models]
 +
* 2024-02: [https://arxiv.org/abs/2408.07055 LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs] ([https://github.com/THUDM/LongWriter code])
 +
* 2024-08: Scientific papers: [The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery]
 +
* 2024-09: PaperQA2: [https://paper.wikicrow.ai/ Language Models Achieve Superhuman Synthesis of Scientific Knowledge] ([https://x.com/SGRodriques/status/1833908643856818443 𝕏 post], [https://github.com/Future-House/paper-qa code])
 +
* 2025-03: [https://arxiv.org/abs/2503.18866 Reasoning to Learn from Latent Thoughts]
 +
* 2025-03: [https://arxiv.org/abs/2503.19065 WikiAutoGen: Towards Multi-Modal Wikipedia-Style Article Generation]
  
 
==Explanation==
 
==Explanation==
Line 84: Line 96:
 
* [https://www.markov.bio/ Markov Bio]: [https://www.markov.bio/research/mech-interp-path-to-e2e-biology Through a Glass Darkly: Mechanistic Interpretability as the Bridge to End-to-End Biology] ([https://x.com/adamlewisgreen/status/1853206279499751531 quick description], [https://markovbio.github.io/biomedical-progress/ background info on recent bio progress])
 
* [https://www.markov.bio/ Markov Bio]: [https://www.markov.bio/research/mech-interp-path-to-e2e-biology Through a Glass Darkly: Mechanistic Interpretability as the Bridge to End-to-End Biology] ([https://x.com/adamlewisgreen/status/1853206279499751531 quick description], [https://markovbio.github.io/biomedical-progress/ background info on recent bio progress])
 
* 2023-01: [https://arxiv.org/abs/2301.05062 Tracr: Compiled Transformers as a Laboratory for Interpretability] ([https://github.com/google-deepmind/tracr code])
 
* 2023-01: [https://arxiv.org/abs/2301.05062 Tracr: Compiled Transformers as a Laboratory for Interpretability] ([https://github.com/google-deepmind/tracr code])
 +
* 2024-10: [https://arxiv.org/abs/2410.03334 An X-Ray Is Worth 15 Features: Sparse Autoencoders for Interpretable Radiology Report Generation]
 
* 2024-12: [https://www.arxiv.org/abs/2412.16247 Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models]
 
* 2024-12: [https://www.arxiv.org/abs/2412.16247 Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models]
 
* 2024-12: [https://arxiv.org/abs/2412.12101 InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders]
 
* 2024-12: [https://arxiv.org/abs/2412.12101 InterPLM: Discovering Interpretable Features in Protein Language Models via Sparse Autoencoders]
 
* 2025-01: [https://arxiv.org/abs/2501.00089 Insights on Galaxy Evolution from Interpretable Sparse Feature Networks]
 
* 2025-01: [https://arxiv.org/abs/2501.00089 Insights on Galaxy Evolution from Interpretable Sparse Feature Networks]
 
* 2025-02: [https://www.biorxiv.org/content/10.1101/2025.02.06.636901v1 From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models]
 
* 2025-02: [https://www.biorxiv.org/content/10.1101/2025.02.06.636901v1 From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models]
 +
* 2025-02: [https://www.goodfire.ai/blog/interpreting-evo-2 Interpreting Evo 2: Arc Institute's Next-Generation Genomic Foundation Model]
  
 
===Uncertainty===
 
===Uncertainty===
Line 139: Line 153:
 
* 2022-11: [https://arxiv.org/abs/2211.09085 Galactica: A Large Language Model for Science]
 
* 2022-11: [https://arxiv.org/abs/2211.09085 Galactica: A Large Language Model for Science]
 
* 2025-03: [https://arxiv.org/abs/2503.17604 OmniScience: A Domain-Specialized LLM for Scientific Reasoning and Discovery]
 
* 2025-03: [https://arxiv.org/abs/2503.17604 OmniScience: A Domain-Specialized LLM for Scientific Reasoning and Discovery]
 +
* 2025-03: Google [https://huggingface.co/collections/google/txgemma-release-67dd92e931c857d15e4d1e87 TxGemma] (2B, 9B, 27B): [https://developers.googleblog.com/en/introducing-txgemma-open-models-improving-therapeutics-development/ drug development]
  
 
=Impact of AI in Science=
 
=Impact of AI in Science=

Latest revision as of 10:40, 29 March 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/ML Methods tailored to Science

Regression (Data Fitting)

Tabular Classification/Regression

Symbolic Regression

Literature Discovery

Commercial

AI/ML Methods in Science

Chemistry

Biology

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

Chemistry

LLMs Optimized for Science

Impact of AI in Science

Related Tools

Literature Search

Data Visualization

Generative

Chemistry

See Also

Retrieved from "http://gisaxs.com/index.php?title=Science_Agents&oldid=7384"