New dataset and multi-agent framework tackle realistic patent drafting with LLMs

Researchers have introduced Dis2Pat, a new dataset designed to train large language models on realistic patent drafting workflows, which typically start with informal inventor disclosures rather than structured legalistic inputs. To address the challenges of generating complete, legally coherent patent applications from such de-legalized materials, they also proposed Patent-MAF, a multi-agent framework for local deployment. Benchmark results indicate that current LLMs struggle with this task, while Patent-MAF establishes a strong baseline, outperforming evaluated open-source models and remaining competitive with larger closed-source models. AI

IMPACT This research could improve the efficiency and accuracy of patent drafting for AI developers and legal professionals.

RANK_REASON The cluster contains a research paper introducing a new dataset and framework for LLM-based patent drafting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Dis2Pat
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• Patent-MAF
• ScienceCast
• scite Smart Citations

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →