New Benchmark Exposes AI's Creative Problem-Solving Weakness

Researchers introduced CresOWLve, a benchmark testing large language models' ability to solve real-world creative puzzles requiring lateral thinking and cross-domain knowledge integration. Testing frontier AI models revealed a significant gap: while LLMs excel at factual retrieval, they struggle to make non-obvious creative connections needed to combine information into solutions—a 17% performance drop from factual to creative tasks.