Crossword Help Forum
Forum Rules

I imagine that entries to the clue-writing competitions on this forum are the product of human ingenuity. This contrasts with 'Winning Move' chess competitions which can instantly be solved by anyone with a chess engine on their computer or smartphone. The Times still offers a prize every Saturday for these.

I am now wondering whether the clue-writing competitions could be spoiled by AI. However, a brief Google comes up with 'Lady Labyrinth' who (which) produces a clue which seems rubbish to me.

Is there any evidence of AI (ChatBot?) having any success with this skill?

As discussed at length here:

https://www.crosswordsolver.org/forum/870710/clueless-629/

ChatGPT is currently crap at solving them, and jono found it's also not very good a writing them..... yet.

Thanks for the link, buddy. I see that I didn't take part in that competition so missed the discussion. However, there were no examples of AI producing its own clue so let's hope it stays that way for a long time.
.

I asked ChatGPT to "write a cryptic crossword clue whose answer is LANDSLIDE" (from Clueless this week).

Attempt #1: Overwhelming victory takes earth by force (8)
Attempt #2: Earth-shaking event grips the hillside (8)

pretty sure neither of these would get a vote.

How about this for nonsense from my own experiment

Command: Write 3 different cryptic clues for 'buddy'

Certainly! Here are three cryptic clues for the word 'buddy':
1. Comrade with no end in sight (5)
Solution: PAL (remove the last letter from "pal")
2. Companion in crime perhaps, always on the run (6)
Solution: AMIGO (an anagram of "crime" with "go" at the end)
3. Close mate, initially hidden in dusk (4)
Solution: CHUM (the initial letters of "close" and "mate" hidden in "dusk")

Command: Write 3 different cryptic clues, whose solution is the word 'buddy'

Certainly! Here are three cryptic clues for the word 'buddy':
1. Clue: Pal found in a hurry (5) Solution: Buddy
Explanation: The word "buddy" is hidden (found) in the phrase "in a hurry."
2. Clue: Comrade missing, but found in buddy (4) Solution: Buddy
Explanation: Remove the letters of "comrade" (missing) from the word "buddy" to get the solution.
3. Clue: Chum wrapped up in buddy film (5) Solution: Buddy Explanation: The word "chum" is reversed (wrapped up) and placed inside the word "buddy" to form the solution.

You might be interested in this PhD thesis from 2007:

https://www.dcs.bbk.ac.uk/site/assets/files/1025/hardcastle.pdf

The author created a system which he called ENIGMA to generate cryptic clues. In a Turing-style test where ENIGMA-generated clues were paired with published clues from newspapers, solvers were asked to identify the human-generated clue. They were correct on 72% of occasions, in other words the computer fooled them 28% of the time.

Appendix E lists the pairs of clues and makes for good reading.

Spotty, interesting. In the clueless thread I referenced below, I wondered what would happen if an LLM was specifically trained on a large database of cryptic clues. Technology has advanced just a hair from 2007.

Examples of some of ENIGMA's better clues. Can you spot the human-authored ones?

SEWER
A. Drain second vessel (5)
B. Second vessel becomes a drain (5)

SCION
A. Twig that coins are lost (5)
B. That is something carrying a charge for a descendant (5)

LAPSE
A. Slight error in peals, possibly (5)
B. Slip plate above boiled peas (5)

SURE
A. Convinced potential user (4)
B. User out, convinced (4)

BROTHER
A. Sibling getting soup with hesitation (7)
B. Double berth is awkward around rising gold (7)

The A clues are all “better” in the sense of more concise and better surface. So unless if was a crummy human setter I’d guess the As are human.

The human-authored clues were all published in the Sun (SURE, LAPSE, BROTHER) or the Indy (SCION, SEWER).