Ask HN: Danger or Fun? encoding secret messages into HN comments
I am a casual observer of HN and am kinda just here as a hobbyist. Most the stuff I read here I can't even comprehend to be honest, you guys speak a lot of gibberish to me.
Ive made something I was surprised is even possible to make which is both fun and daunting in the context of what will be possible in the future.
I asked AI to scrape HN comments into the a large corpus of templates that would be ideal to encode messages in. To my surprise it had nothing preventing it from doing this. It did it instantly. Then I asked it to plug the corpus of templates into this other thing I made which encodes plaintext into small compressed data.
Now I can encode secret messages into HN comments.
Here are a few examples:
> 71 coherent cover texts hiding a secret message https://postimg.cc/mtMmxPkD
and
> 30 cover texts https://postimg.cc/crXVHDwQ
All this really got me thinking, with a sufficient amount of data, you could eventually make something that looks completely believable and relevant to a conversation that is taking place.
How will we know in the future that what we are replying to is not some bot with alter motive? And, how much of this place is already that?
On the topic, what have you guys made that made you think... "Wow, I can't believe I can make this seems potentially dangerous" ? You have rediscovered steganography (https://en.wikipedia.org/wiki/Steganography). Nothing new here on that end. > How will we know in the future that what we are replying to is not some bot with alter motive? How do you know today that you are not doing the same? How did you know last year you were not conversing with some bot? Absent some tell to indicate "likely a bot" you have no way to know with certainty. "On the Internet, nobody knows you're a dog" (https://en.wikipedia.org/wiki/On_the_Internet,_nobody_knows_...) Hmm, so maybe it doesn't even matter. As long as I'm completely blinded to what's going on. Although it could be considered kinda abusive if you thought you were talking to a real person all along. > I asked AI to scrape HN comments Did the AI use the API? (linked at the bottom) https://github.com/HackerNews/API > All this really got me thinking, with a sufficient amount of data, you could eventually make something that looks completely believable and relevant to a conversation that is taking place. I've seen a few similar attempts, and after a while many (hopefully most) are detected, flagged and banned. We already get a ton of bots, but it's still better than other social networks. The only reliable way to get rid of LLM bots is to add friction to community participation, and that's not what we do here, so we get bots, and ones increasingly harder to identify. What's the practical defense to prevent yourself from engaging in fake content? The same way you prevent yourself from engaging with trolls. Be alert, pay attention, when you get the sense you are engaging in fake content (or with a troll) then disengage. Ok, but let's say in the future it becomes impossible to distinguish. I use your comment The same way you [prevent] yourself from [engaging] with trolls. Be alert, [pay attention], when you get the [sense] you are engaging in [fake content] (or with a troll) then [disengage]. The same way you [prohibit] yourself from [involving] with trolls. Be alert, [take note], when you get the [hunch] you are engaging in [fluff] (or with a troll) then [withdrawal]. That's enough data to encode. Does even matter? As long as the replies are sufficiently relevant, and I am completely blinded. I would urge you to be more generous with upvotes, down otes, and article flags. You see something that might be "AI"? Down vote. See an "I wrote a holy book with Claude" article? Flag. Downvote all comments on it. This gets you in the habit of spotting slop, and sends signals to slop posters and HN moderation algorithms. If enough of us do this, we can clean up the joint.