The Download: AI puzzles and a path to our nearest star system

Puzzles and games have long been used to test and improve artificial intelligence (AI) systems.

A notable example is the New York Times Connections puzzle, which groups words into categories based on hidden connections.

In late 2024, the best AI models could solve only 18% of these puzzles, but by early 2025, some models achieved near-perfect scores.

This rapid improvement highlights AI’s growing ability to handle complex, creative tasks.

However, these puzzles also reveal AI’s limitations, as humans often outperform machines in areas requiring intuition or lateral thinking.

The term machine learning originated in 1959 to describe algorithms that improve through experience, such as learning to play checkers.

Chess and Go are other classic AI test cases, where machines have surpassed human champions.

The contrast between AI’s puzzle-solving strengths and weaknesses offers insights into its broader capabilities and gaps.