AI Model Darmok Solves Cryptic Crossword Faster Than Human Champion

AI Achieves Cryptic Crossword Victory

The AI model Darmok recently surpassed human performance in solving a cryptic crossword puzzle. In the final round of the Times's annual cryptic crossword competition, Darmok completed the puzzle approximately three minutes faster than Liam Hughes, the human champion.

Historical Context of AI in Puzzles

This achievement echoes IBM Deep Blue's chess victory over Garry Kasparov in 1997. For decades, experts doubted AI's capacity for complex wordplay, with Will Shortz, the New York Times's puzzle editor, expressing skepticism about computers handling puns and allegories in crosswords. However, earlier AI programs like WebCrow and Dr Fill had already demonstrated success in American-style crosswords, which rely on general knowledge and double meanings.

Darmok's Development and Challenge

Dr. Sid Chatterjee, a computer scientist at the University of Texas in Austin, developed Darmok starting in 2025. The model is named after a Star Trek episode where communication relies on allegories, mirroring the challenge of cryptic crosswords. Darmok's previous attempt at the Times championship ended in failure, but its recent success included solving a clue that stumped multiple human finalists, including a 14-time champion.

Solving Complex Clues

One particular clue, "Wolf trap missing its lowest crosspiece? (5)", demonstrated Darmok's capability. The solution, SNARF, required understanding that "trap" indicated "SNARE" and that removing the bottom line of the 'E' transforms it into an 'F'. This type of visual and wordplay-based clue was previously considered a significant hurdle for AI.