Solving Zebra Puzzles Using Constraint-Guided Multi-Agent Systems
Researchers have developed a new multi-agent system called ZPS that combines Large Language Models (LLMs) with a theorem prover to solve complex logic puzzles like Zebra puzzles. This system breaks down problems, generates code for a theorem prover, and uses agent feedback to refine solutions. In testing, ZPS significantly improved the puzzle-solving accuracy of tested LLMs, with GPT-4 showing a 166% increase in fully correct solutions. AI
IMPACT Introduces a novel approach to enhance LLM reasoning capabilities for complex logical tasks.