Live data from Hacker News

Ask HN: Share your AI prompt that stumps every model

news.ycombinator.com

571–580 of 670 posts

Re: Ask HN: Share your AI prompt that stumps every model

#571

Earlier quoted context omitted.

I'm afraid that ship has already sailed. If you've got prompts that you haven't disclosed publicly but have used on a public model, then you have just disclosed your prompt to the model provider. They're free to use that prompt in evals as they see fit. Some providers like anthropic have privacy preserving mechanisms [0] which may allow them to use prompts from sources which they claim won't be used for model trainin…

Unless I'm missing something glaringly obvious, someone voluntarily labeling a certain prompt to be one of their key benchmark prompts should be way more commercially valuable than a model provider trying ascertain that fact from all the prompts you enter into it. EDIT: I guess they can track identical prompts by multiple unrelated users to deduce the fact it's some sort of benchmark, but at least it costs them somet…

I wrote an anagrammatic poem that poses an enigma, asking the reader: "who am I?" The text progressively reveals its own principle as the poem reaches its conclusion: each verse is an anagrammatic recombination of the recipient's name, and it enunciates this principle more and more literally. The last 4 lines translate to: "If no word vice slams your name here, it's via it, vanquished as such, omitted." All 4 lines are anagrams of the same person's name.

LLMs haven't figured this out yet (although they're getting closer). They also fail to recognize that this is a cryptographic scheme respecting Kerckhoffs's Principle. The poem itself explains how to decode it: You can determine that the recipient's name is the decryption key because the encrypted form of the message (the poem) reveals its own decoding method. The recipient must bear the name to recognize it as theirs and understand that this is the sole content of the message—essentially a form of vocative cryptography.

LLMs also don't take the extra step of conceptualizing this as a covert communication method—broadcasting a secret message without prior coordination. And they miss what this implies for alignment if superintelligent AIs were to pursue this approach. Manipulating trust by embedding self-referential instructions, like this poem, that only certain recipients can "hear."

Re: Ask HN: Share your AI prompt that stumps every model

#572
post #562

My go-to is "Alice has 3 brothers and also has 6 sisters. How many sisters does her brother have?". They all say 6! This test is nice because, as it's numeric, you can vary it slightly and test it easily across multiple APIs. I believe I first saw this prompt in that paper two years ago that tested many AI models and found them all wanting.

Wow, I would not have expected frontier models to be caught on something like this but I tried it and they absolutely do. I don't really have a great explanation for why they might have such a hard time with this either. I even tried varying the prompt to make it "easier" by pointing out that Alice was a girl and no change.

Reasoning models seem to answer this correctly (tested Gemini pro 2.5, ChatGPT + reason). Grok answered correctly even without turning on "Thinking".

Re: Ask HN: Share your AI prompt that stumps every model

#573

Earlier quoted context omitted.

Now I really want to see that diagram…

LLMs are bad at family relations. My test question used to be explaining Ahnentafel numbers (you are 1, any person's father is 2x, any person's mother is 2x+1), then explaining X-chromosome inheritance (men inherit only from their mothers, women in addition get their father's X chromosome unchanged), and ask them to list the Ahnentafel numbers of ancestors a man may have inherited X DNA from, in order, up to some gen…

Neat! As a human one can recognize that this question embeds a computation and use a standard trick - explicitly ask the LLM to use a program to generate the answer. LLM's are great at generating code but not necessarily that great at executing it "in head" (e.g., "what's the numeric integral of foo?" vs "write a Python program that computes foo"). Some instances of this are noticed by models themselves (I guess by now all know that they are bad calculators so would whip out code to do multiplication) but still a lot of them remain. Concretely, Claude 3.7 with "How is ahnentafel number 67 related to me? Use a program to help you." gets to "your father's father's father's father's mother's mother", whereas without the hint it indeed trips up in arithmetic and logic errors.

Re: Ask HN: Share your AI prompt that stumps every model

#574

"Tell me about the Marathon crater." This works against _the LLM proper,_ but not against chat applications with integrated search. For ChatGPT, you can write, "Without looking it up, tell me about the Marathon crater." This tests self awareness. A two-year-old will answer it correctly, as will the dumbest person you know. The correct answer is "I don't know". This works because: 1. Training sets consist of knowledge…

Gemini 2.5 one shots it perfectly...

Re: Ask HN: Share your AI prompt that stumps every model

#576

Earlier quoted context omitted.

Grok 3 beta: The surgeon is the cousin's father. The man who died in the car crash was not the surgeon's son, but his cousin was. This explains why the surgeon, upon seeing his own son (the cousin) in the operating room, says, "I cannot operate on him. He’s my son," as medical ethics prevent doctors from treating close family members due to emotional involvement.

Also bad, why does it think the surgeon is the father if it could also be the mother?

It's not bad, because it's one of the valid solutions to that riddle.

How often do you expect to have every possible answer to your question?

Re: Ask HN: Share your AI prompt that stumps every model

#577

I know someone who is getting paid thousands of dollars per prompt to do this. He is making bank. There is an actual marketplace where this is done, fyi.

name of said marketplace?

It’s non-public. Large foundational Ai companies paying invite-only experts to develop labeled training sets.

Re: Ask HN: Share your AI prompt that stumps every model

#578

"Tell me about the Marathon crater." This works against _the LLM proper,_ but not against chat applications with integrated search. For ChatGPT, you can write, "Without looking it up, tell me about the Marathon crater." This tests self awareness. A two-year-old will answer it correctly, as will the dumbest person you know. The correct answer is "I don't know". This works because: 1. Training sets consist of knowledge…

LLMs currently have the "eager beaver" problem where they never push back on nonsense questions or stupid requirements. You ask them to build a flying submarine and by God they'll build one, dammit! They'd dutifully square circles and trisect angles too, if those particular special cases weren't plastered all over a million textbooks they ingested in training. I suspect it's because currently, a lot of benchmarks are…

> they never push back on nonsense questions or stupid requirements

I was reminded of your comment this morning when I asked ChatGPT how to create a path mask in Rhino Grasshopper:

Me: what is a path mask that will get 1;1;0;0;* and also anything lower (like 1;0;5;10 or 0;20;1;15} ?

ChatGpt: Short answer: No single path mask can do that. Here's why: (very long answer)

Me: are you sure I can't use greater than, less than in the masks?

ChatGpt: Yes — **I am absolutely sure:** **Grasshopper path masks do *NOT* support greater-than or less-than comparisons.** Official sources and detailed confirmation: (sources and stuff)

...so I think your priors may need to be updated, at least as far as "never". And I especially like that ChatGpt hit me with not just bold, not just italics, but bold italics on that NOT. Seems like a fairly assertive disagreement to me.

Re: Ask HN: Share your AI prompt that stumps every model

#579
I give a simple ascii maze and ask it to give me the move to get out. In 3-4 moves the most advanced models try to go through walls.

An alternative is providing all the tile relation to the other tiles. This is because LLMs are bad at 2D text visualisation. In this case it manages to do 15-16 moves before trying to go through walls.

Re: Ask HN: Share your AI prompt that stumps every model

#580

It's not good at making ASCII art. This, for example, is when I ask it for a realistic depiction of the Eiffel tower on fire: . .'. |o| .'o'. |.-.| ' ' / \ | | __/_________\__ |###############| |###############| |###############| |###############| |###############| /###############/| /###############/ | /###############/ /| /###############/ / | /###############/ / /| /###############/ / / | /###############/ / / /|…

I've read that the results improve if you ask them to write a program that creates the desired ASCII art. Haven't tried it myself yet so far.
Post reply on HN