Can You Solve Connections Better Than a Computer?
A research team led by Associate Professor Julian Togelius challenged a variety of modern natural language processing systems to solve The New York Times' daily puzzle Connections. In the study, AI struggled to complete the puzzle, with GPT-4, the most successful of the models, only solving about 29 percent of the time and performing the worst on "tricky" word associations, just like humans. When GPT-4 was given step-by-step prompts to guide it through the reasoning of the puzzle, its performance was slightly better, at 39 percent.
Researchers
