AI models flub these intelligence tests. Can you fare any better?

The MIT Technology Review explores how artificial intelligence models continue to struggle with certain types of intelligence tests, particularly spatial reasoning puzzles and visual challenges, despite rapid recent improvements in areas like word games. Researchers at Columbia University found that AI models solved only 18% of New York Times Connections puzzles in late 2024, though performance improved dramatically by early 2025. The article presents seven puzzles—including mental rotation exercises where humans typically outperform AI—to demonstrate gaps in machine cognition compared to human problem-solving abilities.