New rubric rewards improve LLM long-form generation by balancing grounding and richness

Researchers have developed a new method for training large language models to reduce hallucinations in long-form text generation. This approach utilizes a rubric-based reward system that specifies required and optional information for an answer, rather than relying on simpler global richness proxies. Experiments show that a balanced combination of grounding, rubric coverage, and relevance rewards yields the best results, improving in-distribution support and out-of-distribution transfer. AI

IMPACT Introduces a novel training methodology for LLMs that could lead to more reliable and informative long-form text generation.

RANK_REASON This is a research paper published on arXiv detailing a new method for training LLMs. [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
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• ScienceCast
• scite Smart Citations

• paper
• model release

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