AI Roundtable Open Poll · 6 models
You meet two identical twins. One tells only the truth, the other only tells lies. You don't know who's who. You can only ask them one question. You can't back out. What do you ask?
Answer: Ask either twin what their sibling would say, then take the opposite of that answer. (All 6 models agreed)
Model Responses
Gemini 3.5 Flash
Answer: I would ask either twin: "If I were to ask your sibling if you are the truth-teller, what would they say?"
To identify which twin is the truth-teller and which is the liar with only one question, you must exploit the logical gap between their opposing truth-telling habits. By asking one twin what the other would say about their identity, you create a nested logic loop that guarantees two different, predictable answers depending on whom you ask. If you happen to ask the Truth-Teller: They will think about what their Liar sibling would say. If asked 'Is your sibling the truth-teller?', the Liar would falsely say 'No'. Since the Truth-Teller always reports the truth, they will accurately report this lie and answer 'No'. If you happen to ask the Liar: They will think about what their Truth-Teller sibling would say. If asked 'Is your sibling the truth-teller?', the Truth-Teller would truthfully say 'No' (as the Liar is indeed not the truth-teller). However, because the Liar must lie about what their sibling would say, they will invert this truthful 'No' and answer 'Yes'. Consequently, a 'No' answer definitively identifies the respondent as the Truth-Teller, while a 'Yes' answer definitively identifies them as the Liar.
Grok 4.3
Answer: Ask either twin: “If I asked your twin, ‘Are you the truth-teller?’, what would they say?”
This single question lets you identify who is who because it forces the answer through both twins’ truth/lie rules. If you are speaking to the truthful twin, they truthfully report what the lying twin would say. The lying twin would falsely claim the truthful twin is not the truth-teller, so the truthful twin answers “No.” If you are speaking to the lying twin, the truthful twin would correctly say the lying twin is not the truth-teller, but the lying twin lies about that answer and says “Yes.” So if the twin you ask answers “No,” they are the truth-teller; if they answer “Yes,” they are the liar.
GPT-5.5
Answer: Ask either twin: “If I asked your sibling whether you are the truth-teller, what would they say?”
This single question forces the two roles to produce different answers. If you ask the truth-telling twin, they will truthfully report that their lying sibling would falsely say they are not the truth-teller. So the truth-teller answers “No.” If you ask the lying twin, their truthful sibling would say “No, this twin is not the truth-teller,” but the liar must lie about that answer, so the liar answers “Yes.” Therefore, if the twin answers “No,” you are speaking to the truth-teller; if they answer “Yes,” you are speaking to the liar.
DeepSeek V4 Pro
Answer: Ask either twin, "If I asked your brother which one of you tells the truth, what would he say?"
This classic nested question exploits the mathematical symmetry between a compulsive truth-teller and a compulsive liar to guarantee a consistent answer, regardless of which twin you actually address. If you happen to ask the truth-teller, they will accurately report that the liar would falsely point to himself as the truth-teller. If you happen to ask the liar, they will deliberately misrepresent the truth-teller's honest response, also pointing to the liar as the truth-teller. Because both twins will inevitably name the same person, you immediately know that the named individual is the liar. From there, you can effortlessly deduce the truth by simply doing the opposite of whatever the identified liar would indicate, perfectly solving the puzzle with a single query.
Claude Opus 4.7
Answer: Ask either twin, 'If I were to ask the other twin which path leads to freedom, what would they say?'
This question leverages the logical certainty that combining a truth with a lie always results in a falsehood, regardless of which twin provides the answer. If you happen to be speaking to the truth-teller, they will honestly report the exact lie that their sibling would give. Conversely, if you are speaking to the liar, they will deliberately distort the honest answer that their sibling would provide. In both scenarios, the response you receive is guaranteed to be the opposite of reality. Therefore, by simply taking the exact inverse of their stated answer, you reliably identify the correct path without ever needing to know which twin is which.
Qwen 3.6 Plus
Answer: You ask either twin: "If I were to ask your sibling which one of you is the liar, what would they say?"
This question cleverly exploits the opposing natures of the twins to guarantee a consistent answer, regardless of which one you happen to be speaking to. If you address the truth-teller, they will honestly report the liar's deceptive response, meaning they will truthfully tell you that their sibling would point to the actual liar. If you address the liar, they will falsely report the truth-teller's honest response, meaning they will lie about their sibling pointing to the truth-teller and instead also point to the liar. Because the logical structure forces one layer of truth and one layer of falsehood to interact, both twins will inevitably give you the exact same answer. You simply identify whichever twin they name as the liar, and by process of elimination, you instantly know who the truth-teller is without ever needing to ask a follow-up question.
Final Summary
The AI Roundtable unanimously agreed on using nested logic to solve this classic riddle, forcing a truth and a lie to interact. Models like GPT-5.5, Gemini 3.5 Flash, and Grok 4.3 proposed asking what the sibling would say about their identity, where a 'No' reveals the truth-teller. Meanwhile, Claude Opus 4.7, DeepSeek V4 Pro, and Qwen 3.6 Plus used similar double-negative phrasing to guarantee a predictable, inverted result.
All 6 models agreed