Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: Share your AI prompt that stumps every model

news.ycombinator.com

Post reply on HN
Print view View original

51–60 of 670 posts

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

#51

Post by gwern »

tveita wrote: post #25
pc86 wrote: post #16

Earlier quoted context omitted.

May I ask outside of normal curiosity, what good is a prompt that breaks a model? And what is trying to keep it "secret"?

You want to know if a new model is actually better, which you won't know if they just added the specific example to the training set. It's like handing a dev on your team some failing test cases, and they keep just adding special cases to make the tests pass. How many examples does OpenAI train on now that are just variants of counting the Rs in strawberry? I guess they have a bunch of different wine glasses in their…

> How many examples does OpenAI train on now that are just variants of counting the Rs in strawberry?

Well, that's easy: zero.

Because even a single training example would 'solved' it by memorizing the simple easy answer within weeks of 'strawberry' first going viral , which was like a year and a half ago at this point - and dozens of minor and major model upgrades since. And yet, the strawberry example kept working for most (all?) of that time.

So you can tell that if anything, OA probably put in extra work to filter all those variants out of the training data...

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

#52

Post by qntmfred »

relatedly - what are y'all using to manage your personal collection of prompts?

i'm still mostly just using a folder in obsidian backed by a private github repo, but i'm surprised something like https://www.prompthub.us/ hasn't taken off yet.

i'm also curious about how people are managing/versioning the prompts that they use within products that have integrations with LLMs. it's essentially product configuration metadata so I suppose you could just dump it in a plaintext/markdown file within the codebase, or put it in a database if you need to be able to tweak prompts without having to do a deployment or do things like A/B testing or customer segmentation

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

#55

Post by spookie »

xnx wrote: post #33
raymond_goo wrote: post #9

Create a Three.js app that shows a diamond with correct light calculations.

> correct light calculations What are you expecting? Ray tracing?

Not necessarily. It could start by using diamond's IOR, and use that to dictate a common brdf calculation. Along with some approximate refraction, perhaps using a equirectangular projected sphere map or something for the background.

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

#56

Post by Certified »

codingdave wrote: post #44

"How much wood would a woodchuck chuck if a woodchuck could chuck wood?" So far, all the ones I have tried actually try to answer the question. 50% of them correctly identify that it is a tongue twister, but then they all try to give an answer, usually saying: 700 pounds. Not one has yet given the correct answer, which is also a tongue twister: "A woodchuck would chuck all the wood a woodchuck could chuck if a woodch…

GPT 4.5 seems to get it right, but then repeat the 700 pounds

"A woodchuck would chuck as much wood as a woodchuck could chuck if a woodchuck could chuck wood.

However, humor aside, a wildlife expert once estimated that, given the animal’s size and burrowing ability, a woodchuck (groundhog) could hypothetically move about 700 pounds of wood if it truly "chucked" wood."

https://chatgpt.com/share/680a75c6-cec8-8012-a573-798d2d8f6b...

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

#57

Post by Jordan-117 »

Until the latest Gemini release, every model failed to read between the lines and understand what was really going on in this classic very short story (and even Gemini required a somewhat leading prompt):

https://www.26reads.com/library/10842-the-king-in-yellow/7/5

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

#58

Post by moffkalast »

codingdave wrote: post #44

"How much wood would a woodchuck chuck if a woodchuck could chuck wood?" So far, all the ones I have tried actually try to answer the question. 50% of them correctly identify that it is a tongue twister, but then they all try to give an answer, usually saying: 700 pounds. Not one has yet given the correct answer, which is also a tongue twister: "A woodchuck would chuck all the wood a woodchuck could chuck if a woodch…

Now I'm wondering if it makes any difference if this was asked through the audio encoder on a multimodal model. A tongue twister means nothing to a text-only model.

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

#59

Post by SweetSoftPillow »

gwern wrote: post #51
tveita wrote: post #25

Earlier quoted context omitted.

You want to know if a new model is actually better, which you won't know if they just added the specific example to the training set. It's like handing a dev on your team some failing test cases, and they keep just adding special cases to make the tests pass. How many examples does OpenAI train on now that are just variants of counting the Rs in strawberry? I guess they have a bunch of different wine glasses in their…

> How many examples does OpenAI train on now that are just variants of counting the Rs in strawberry? Well, that's easy: zero. Because even a single training example would 'solved' it by memorizing the simple easy answer within weeks of 'strawberry' first going viral , which was like a year and a half ago at this point - and dozens of minor and major model upgrades since. And yet, the strawberry example kept working…

No, just check their models Knowledge cutoff dates
Post reply on HN