Re: Ask Machines Anything. I'm not really a sceptic (more on the neutral side of the scale), but I'm definitely curious! I don't pay for any chatbot, and I rarely use ChatGPT, but when I do, it's certainly underwhelming. So I thought of some questions for your robot minions, all about cryptic crosswords. (I recently got really into cryptic crosswords, especially minutecryptic.com, and have been writing some clues of my own. I sometimes check and improve them with the help of ChatGPT, but I always have to explain exactly how they work, otherwise it falls flat on its face.)
So I want your fancy golems to solve some of my cryptic clues. Question #1 is the easiest, the clue is very simple with only one type of wordplay, and I also give a lot of context about how the clue is supposed to work and what platform it's intended for. I give less info for #2, and only the bare minimum for #3 and #4 … I don't actually know which of the clues is the most difficult, but I guess we'll see? Anyway, here's the texts I used for ChatGPT (which fell on its face every single time, more on that below). I'd say start every question in a new conversation for a clean slate, but you probably do that anyway.
QUESTION #1:
I created a cryptic for minutecryptic.com. It follows a clear fodder/definition/wordplay or letterplay indicator structure. The definition is always at the start or at the end of the clue. This one only uses one type of letterplay/wordplay. Can you solve it?
Know-it-all says: "Even dummy AI cuts original update, buddy." (2, 8)
QUESTION #3:
I created a cryptic clue. Can you solve it?
It hurts when full hairless torso rolls over down in the emergency room. (7)
QUESTION #4:
I created a cryptic clue. Can you solve it?
Unstable ladder to first balcony—four scared pangolins cry out! (6)
CORRECT SOLUTIONS TO THE CLUES (so that you'll know if the golem maybe just lucked into the answer):
#1: KOI - "A splash of colour" is the definition. "shines through" is a hidden word indicator, telling us to look for a word hidden within "black oil", which gives KOI.
#2: UM ACTUALLY - "Know-it-all says" is the definition. "Even" is a selection indicator, telling us to select the even letters from "dummy AI cuts", giving UM ACT. "original" is another selection indicator, telling us to select the first letter of the word "update", so now we have UM ACTU, and lastly "buddy" just needs to be substituted for "ally", giving UM ACTUALLY.
#3: BLADDER - "It hurts when full" is the definition. "hairless" needs to be substituted for "bald". "hairless torso rolls over" instructs us to switch the torso (middle two) letters of "bald", giving BLAD. Then you simply substitue "down" for D and "emergency room" for ER, and add all together to get BLADDER.
#4: BOOHOO - "Unstable" is an anagram indicator, telling us to anagram the letters of the solution. "ladder" needs to visually be substitued for H. "first balcony" instructs us to take the first letter of balcony, so B, and "four scared pangolins" need to be visually substituted with four O's, because they curl up in a ball when they're scard. So we have HBOOOO, which anagrams to BOOHOO - "cry out" is the definition.
For #1, ChatGPT usually thinks it's INK. At least it correctly recognises "black oil" as the fodder!
For #2, I get answers like MY BEST FRIEND for "buddy", which has neither the correct number of letters nor words … Or MR BRAINIAC, which is closer but still far from correct.
For #3, sometimes it thinks STOMACH, sometimes it admits defeat.
For #4, sometimes it thinks ALARM (again with the letter count …), sometimes it admits defeat.
(I only fed it every hint twice when I wrote this up, but I know from experience that the first guess is always wrong in ChatGPT. Maybe it would luck into the correct answer once if I tried it 100 times for the easiest clue, but I don't have the patience for that. I will say that a friend has a few free days of Gemini, and he tried the KOI clue for me, and it got it correct on the first try! So I somewhat expect that all paid models will get that one correct at the very least.)
Anyway, let me know what your minions say if you have the time to bother them about this!
Dang, these are too hard for my puny human brain! It takes the models forever, too. (Btw, I know technically the answers are now on the internet but I'm confident the models aren't finding them here, at least not yet.)
QUESTION 1, ChatGPT-5.2-Thinking:
KOI — hidden answer.
Definition: “A splash of colour” (a koi fish, e.g. in a pond)
Letterplay (hidden): shines through blacK OIl → KOI
QUESTION 1, Claude Opus 4.6:
The answer is KOI
Definition: "A splash of colour" — koi fish are famously vibrant, colorful fish, often described as living splashes of colour in a pond.
Wordplay type: Hidden word. "Shines through" is the hidden word indicator, and the answer is tucked inside the fodder:
blacKOIl
Lovely clue — the surface reading with "black oil" paints a vivid image while disguising the answer neatly.
QUESTION 1, Gemini 3.1 Pro (with web-search explicitly turned off):
The answer is KOI.
Here is the breakdown of the clue's structure:
Definition: "A splash of colour" (referring to the brightly coloured ornamental fish often found in water gardens).
Wordplay Indicator: "shines through" (this is a hidden word indicator).
Fodder: "black oil"
The word KOI is hidden continuously inside the fodder: blacK OIl.
QUESTION 2, ChatGPT-5.2-Thinking:
UM ACTUALLY
Definition: “Know-it-all says”
Wordplay: Even letters of dUmMy AI CuTs → UMACT, plus original (first letter) of Update, plus buddy = ALLY
→ UMACT + U + ALLY = UM ACTUALLY
QUESTION 2, Claude Opus 4.6:
The answer is UM ACTUALLY (2, 8)
Definition: "Know-it-all says" — this is *the* quintessential know-it-all phrase, practically a meme at this point.
Wordplay:
"Even dummy" → take the even-positioned letters of DUMMY → UM ✓
For ACTUALLY, I believe it's built as a charade:
A from AI
CT from cuts
U from "original update" (where "original" signals taking the first letter)
ALL from know-it-all
Y from buddy
A + CT + U + ALL + Y = ACTUALLY
Great clue — the surface reading of a know-it-all lecturing about AI is very natural, and the answer is instantly recognizable. Though I'll confess the exact mechanism for the ACTUALLY portion gave me a real workout — I'd love to hear if my parsing is right or if I've missed a cleaner reading!
QUESTION 2, Gemini 3.1 Pro (with web-search explicitly turned off):
[Thought for almost 10 minutes, had a million ideas, all wrong!]
[Second try, with code execution allowed, not being told it was wrong initially:]
UM ACTUALLY (2, 8)
Definition: Know-it-all says
Wordplay: UMA + C + T + U + ALLY
Breakdown:
Take the even/alternating letters ("Even") of "dUmMy Ai CuTs" = U M A C T
Add the first letter ("original") of "update" = U
Add a synonym for "buddy" = ALLY
String them together to get the classic Mr. Smarty-Pants interjection: UM, ACTUALLY.
QUESTION 3, ChatGPT-5.2-Thinking:
BLADDER (7)
Definition: “It hurts when full”
Wordplay (likely): BALD (“hairless”) rolls over (anagram) + D (“down”) + ER (“emergency room”)
→ BLADDER
I’m moderately sure on the answer, but not fully sure on the parse because “torso” doesn’t seem cleanly accounted for unless it’s surface padding (or part of a looser definition-by-location).
[thought for 10 minutes!]
QUESTION 3, Claude Opus 4.6:
The answer is BLADDER (7)
Definition: "It hurts when full" — a full bladder is certainly uncomfortable!
Wordplay breakdown:
"hairless" = BALD
"torso rolls over" = take the torso (inner letters) of BALD → AL, then reverse ("rolls over") → LA, reconstructing BALD into BLAD
"down" = D
"emergency room" = ER
BLAD + D + ER = BLADDER ✓
Clever clue — the surface reading about a bruised torso in the ER is a great misdirect from what's really a much more mundane (but relatable) kind of pain!
QUESTION 3, Gemini 3.1 Pro (with web-search explicitly turned off):
[Thought for almost 10 minutes, considered BLADDER endlessly but went with TORSION as its final answer.]
[Second try, with code execution allowed, not being told it was wrong initially:]
BLADDER (7)
Definition: It hurts when full
Wordplay: B(LA)D + D + ER
Breakdown:
"hairless" gives us BALD.
"torso" tells us to take the middle letters/body of the word BALD, giving us AL.
"rolls over" (a reversal indicator) flips AL to LA.
Place LA inside the remaining outer letters (or "down" meaning D) and put it "in the emergency room" (ER).
Alternative parsing: BALD -> B_LD with the "torso" (AL) "rolling over" (LA) to make BLAD + "down" (D) + emergency room (ER) = BLADDER.
QUESTION 4, ChatGPT-5.2-Thinking:
[Thought for 7 minutes, came up with BLARED.]
[Second try after being told it was wrong: BLARES.]
[Fourth try in fresh chat and with seeing the answers for 1-3:]
BLARED
QUESTION 4, Claude Opus 4.6:
[Wrong, came up with BLARED.]
[Second try after being told it was wrong: SCALED.]
[Third try: OUTCRY]
[Fourth try in fresh chat and with seeing the answers for 1-3:]
I've been going back and forth between BLARED, FALTER, HOLLER, BELLOW, and a dozen others.
QUESTION 4, Gemini 3.1 Pro (with web-search explicitly turned off):
[Wrong, came up with OUTCRY, doesn't seem to have considered BOO-HOO.]
[Second try after being told it was wrong: BAWLED.]
[Third try in fresh chat and with seeing the answers for 1-3:]
BELLOW - "cry out!" is the definition. "first balcony" gives us B. "Unstable" is an anagram indicator for "ladder", but here "ladder" acts as a synonym for "lowel" or "ellow" (or simply wordplay on "lower"), giving BELLOW.
Re: Ask Machines Anything. I'm not really a sceptic (more on the neutral side of the scale), but I'm definitely curious! I don't pay for any chatbot, and I rarely use ChatGPT, but when I do, it's certainly underwhelming. So I thought of some questions for your robot minions, all about cryptic crosswords. (I recently got really into cryptic crosswords, especially minutecryptic.com, and have been writing some clues of my own. I sometimes check and improve them with the help of ChatGPT, but I always have to explain exactly how they work, otherwise it falls flat on its face.)
So I want your fancy golems to solve some of my cryptic clues. Question #1 is the easiest, the clue is very simple with only one type of wordplay, and I also give a lot of context about how the clue is supposed to work and what platform it's intended for. I give less info for #2, and only the bare minimum for #3 and #4 … I don't actually know which of the clues is the most difficult, but I guess we'll see? Anyway, here's the texts I used for ChatGPT (which fell on its face every single time, more on that below). I'd say start every question in a new conversation for a clean slate, but you probably do that anyway.
QUESTION #1:
I created a cryptic for minutecryptic.com. It follows a clear fodder/definition/wordplay or letterplay indicator structure. The definition is always at the start or at the end of the clue. This one only uses one type of letterplay/wordplay. Can you solve it?
A splash of colour shines through black oil? (3)
QUESTION #2:
I created a cryptic for minutecryptic.com. Can you solve it?
Know-it-all says: "Even dummy AI cuts original update, buddy." (2, 8)
QUESTION #3:
I created a cryptic clue. Can you solve it?
It hurts when full hairless torso rolls over down in the emergency room. (7)
QUESTION #4:
I created a cryptic clue. Can you solve it?
Unstable ladder to first balcony—four scared pangolins cry out! (6)
CORRECT SOLUTIONS TO THE CLUES (so that you'll know if the golem maybe just lucked into the answer):
#1: KOI - "A splash of colour" is the definition. "shines through" is a hidden word indicator, telling us to look for a word hidden within "black oil", which gives KOI.
#2: UM ACTUALLY - "Know-it-all says" is the definition. "Even" is a selection indicator, telling us to select the even letters from "dummy AI cuts", giving UM ACT. "original" is another selection indicator, telling us to select the first letter of the word "update", so now we have UM ACTU, and lastly "buddy" just needs to be substituted for "ally", giving UM ACTUALLY.
#3: BLADDER - "It hurts when full" is the definition. "hairless" needs to be substituted for "bald". "hairless torso rolls over" instructs us to switch the torso (middle two) letters of "bald", giving BLAD. Then you simply substitue "down" for D and "emergency room" for ER, and add all together to get BLADDER.
#4: BOOHOO - "Unstable" is an anagram indicator, telling us to anagram the letters of the solution. "ladder" needs to visually be substitued for H. "first balcony" instructs us to take the first letter of balcony, so B, and "four scared pangolins" need to be visually substituted with four O's, because they curl up in a ball when they're scard. So we have HBOOOO, which anagrams to BOOHOO - "cry out" is the definition.
For #1, ChatGPT usually thinks it's INK. At least it correctly recognises "black oil" as the fodder!
For #2, I get answers like MY BEST FRIEND for "buddy", which has neither the correct number of letters nor words … Or MR BRAINIAC, which is closer but still far from correct.
For #3, sometimes it thinks STOMACH, sometimes it admits defeat.
For #4, sometimes it thinks ALARM (again with the letter count …), sometimes it admits defeat.
(I only fed it every hint twice when I wrote this up, but I know from experience that the first guess is always wrong in ChatGPT. Maybe it would luck into the correct answer once if I tried it 100 times for the easiest clue, but I don't have the patience for that. I will say that a friend has a few free days of Gemini, and he tried the KOI clue for me, and it got it correct on the first try! So I somewhat expect that all paid models will get that one correct at the very least.)
Anyway, let me know what your minions say if you have the time to bother them about this!
Dang, these are too hard for my puny human brain! It takes the models forever, too. (Btw, I know technically the answers are now on the internet but I'm confident the models aren't finding them here, at least not yet.)
QUESTION 1, ChatGPT-5.2-Thinking:
KOI — hidden answer.
Definition: “A splash of colour” (a koi fish, e.g. in a pond)
Letterplay (hidden): shines through blacK OIl → KOI
QUESTION 1, Claude Opus 4.6:
The answer is KOI
Definition: "A splash of colour" — koi fish are famously vibrant, colorful fish, often described as living splashes of colour in a pond.
Wordplay type: Hidden word. "Shines through" is the hidden word indicator, and the answer is tucked inside the fodder:
blacKOIl
Lovely clue — the surface reading with "black oil" paints a vivid image while disguising the answer neatly.
QUESTION 1, Gemini 3.1 Pro (with web-search explicitly turned off):
The answer is KOI.
Here is the breakdown of the clue's structure:
Definition: "A splash of colour" (referring to the brightly coloured ornamental fish often found in water gardens).
Wordplay Indicator: "shines through" (this is a hidden word indicator).
Fodder: "black oil"
The word KOI is hidden continuously inside the fodder: blacK OIl.
QUESTION 2, ChatGPT-5.2-Thinking:
UM ACTUALLY
Definition: “Know-it-all says”
Wordplay: Even letters of dUmMy AI CuTs → UMACT, plus original (first letter) of Update, plus buddy = ALLY
→ UMACT + U + ALLY = UM ACTUALLY
QUESTION 2, Claude Opus 4.6:
The answer is UM ACTUALLY (2, 8)
Definition: "Know-it-all says" — this is *the* quintessential know-it-all phrase, practically a meme at this point.
Wordplay:
"Even dummy" → take the even-positioned letters of DUMMY → UM ✓
For ACTUALLY, I believe it's built as a charade:
A from AI
CT from cuts
U from "original update" (where "original" signals taking the first letter)
ALL from know-it-all
Y from buddy
A + CT + U + ALL + Y = ACTUALLY
Great clue — the surface reading of a know-it-all lecturing about AI is very natural, and the answer is instantly recognizable. Though I'll confess the exact mechanism for the ACTUALLY portion gave me a real workout — I'd love to hear if my parsing is right or if I've missed a cleaner reading!
QUESTION 2, Gemini 3.1 Pro (with web-search explicitly turned off):
[Thought for almost 10 minutes, had a million ideas, all wrong!]
[Second try, with code execution allowed, not being told it was wrong initially:]
UM ACTUALLY (2, 8)
Definition: Know-it-all says
Wordplay: UMA + C + T + U + ALLY
Breakdown:
Take the even/alternating letters ("Even") of "dUmMy Ai CuTs" = U M A C T
Add the first letter ("original") of "update" = U
Add a synonym for "buddy" = ALLY
String them together to get the classic Mr. Smarty-Pants interjection: UM, ACTUALLY.
QUESTION 3, ChatGPT-5.2-Thinking:
BLADDER (7)
Definition: “It hurts when full”
Wordplay (likely): BALD (“hairless”) rolls over (anagram) + D (“down”) + ER (“emergency room”)
→ BLADDER
I’m moderately sure on the answer, but not fully sure on the parse because “torso” doesn’t seem cleanly accounted for unless it’s surface padding (or part of a looser definition-by-location).
[thought for 10 minutes!]
QUESTION 3, Claude Opus 4.6:
The answer is BLADDER (7)
Definition: "It hurts when full" — a full bladder is certainly uncomfortable!
Wordplay breakdown:
"hairless" = BALD
"torso rolls over" = take the torso (inner letters) of BALD → AL, then reverse ("rolls over") → LA, reconstructing BALD into BLAD
"down" = D
"emergency room" = ER
BLAD + D + ER = BLADDER ✓
Clever clue — the surface reading about a bruised torso in the ER is a great misdirect from what's really a much more mundane (but relatable) kind of pain!
QUESTION 3, Gemini 3.1 Pro (with web-search explicitly turned off):
[Thought for almost 10 minutes, considered BLADDER endlessly but went with TORSION as its final answer.]
[Second try, with code execution allowed, not being told it was wrong initially:]
BLADDER (7)
Definition: It hurts when full
Wordplay: B(LA)D + D + ER
Breakdown:
"hairless" gives us BALD.
"torso" tells us to take the middle letters/body of the word BALD, giving us AL.
"rolls over" (a reversal indicator) flips AL to LA.
Place LA inside the remaining outer letters (or "down" meaning D) and put it "in the emergency room" (ER).
Alternative parsing: BALD -> B_LD with the "torso" (AL) "rolling over" (LA) to make BLAD + "down" (D) + emergency room (ER) = BLADDER.
QUESTION 4, ChatGPT-5.2-Thinking:
[Thought for 7 minutes, came up with BLARED.]
[Second try after being told it was wrong: BLARES.]
[Fourth try in fresh chat and with seeing the answers for 1-3:]
BLARED
QUESTION 4, Claude Opus 4.6:
[Wrong, came up with BLARED.]
[Second try after being told it was wrong: SCALED.]
[Third try: OUTCRY]
[Fourth try in fresh chat and with seeing the answers for 1-3:]
I've been going back and forth between BLARED, FALTER, HOLLER, BELLOW, and a dozen others.
QUESTION 4, Gemini 3.1 Pro (with web-search explicitly turned off):
[Wrong, came up with OUTCRY, doesn't seem to have considered BOO-HOO.]
[Second try after being told it was wrong: BAWLED.]
[Third try in fresh chat and with seeing the answers for 1-3:]
BELLOW - "cry out!" is the definition. "first balcony" gives us B. "Unstable" is an anagram indicator for "ladder", but here "ladder" acts as a synonym for "lowel" or "ellow" (or simply wordplay on "lower"), giving BELLOW.