Designed to confound AI, the ARC Challenge puzzles are now puzzling thousands of humans as well.
Created by Google engineer François Cholet, the ARC benchmark (Abstraction and Reasoning Corpus) is a set of small puzzles that test how well AI models can solve problems based on a set of examples. These puzzles are designed to check whether the models can understand patterns and figure out solutions from limited information. Researchers can see whether AI systems can think creatively and solve problems like humans do.
Chollet, an artificial intelligence researcher and senior staff software engineer at Google, announced the ARC Prize, a $1 million competition to create an AI that can adapt to new things and solve simple reasoning problems.
In an X post announcing the ARC award, Shollett said that the ARC benchmark was launched four years ago: “ARC tasks are easy for humans. They’re not complex. They require no expertise. Even a child can solve them. But modern AI struggles with them,” he wrote.
But one of the puzzles on the ARC Prize website has left thousands of social media users stumped, with screenshots of the puzzle being widely shared on X.
Check out the ARC Challenge Puzzle below.
“I showed this to my dad and he couldn’t solve it. He’s 57 years old,” User X wrote while sharing the puzzle. His post has been viewed over 1.8 million times.
Got it? If not, here’s a hint: The grey boxes in the “Input” section are separators and are not part of the puzzle.
solution
Several X users have explained that the red square in the “Output” represents where the two blue patterns intersect, and the grey line is the border between the two blue patterns in the “Input” section.
When you overlay the two blue patterns, the overlapping rectangles are the “output” solution.
Here’s a more visual explanation:
For more information on similar issues, see the ARC Prize website.
In the X thread announcing the ARC prize, Chollet explained that ARC problems are difficult for AI because they are designed to be memorizable.
