Frontier LLMs Distill Answer Set Programming Theories with High Accuracy

A new study explored distilling Answer Set Programming (ASP) theories from large language models using a neurosymbolic approach. The research tested nine models, including frontier models like Claude Sonnet 4.6, Claude Opus 4.7, GPT-5, and DeepSeek V4 Pro, across three benchmarks: CLEVR, GQA, and CLEVRER. Most frontier models achieved high accuracy, with Sonnet, Opus, and DeepSeek V4 Pro reaching over 90% on CLEVRER. GPT-5 performed exceptionally on CLEVR but struggled with GQA and CLEVRER, with its accuracy decreasing when reference theories were introduced. AI

IMPACT This research demonstrates LLMs' potential in neurosymbolic reasoning, potentially advancing automated theory generation for complex problem-solving.

RANK_REASON The item describes a research paper detailing experiments with LLMs on a specific task (distilling ASP theories) and reporting benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• answer set programming
• Claude Opus 4.7
• Claude Sonnet 4.6
• CLEVR
• CLEVRER
• DeepSeek-V4 Flash
• DeepSeek V4 Pro
• GPT-5
• GPT-OSS 120B
• GPT OSS 20B
• GQA
• Hugging Face
• Qwen3.5:9b
• Qwen3.6-27B

• paper
• model release

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