Hacker News Reader

Comment by liorben-david | Hacker News Reader

This is awesome!

Game idea to build on top of this: Table top deception type game where each agent has the goal of convincing the real users that they are in fact also real users.(So each agent is trying to pass a turing test).

Every AI agent uses RL to optimally prompt their personal LLM for how they should chat with the human players. eg should they try to frame a certain person, should they play it dumb, should they gaslight etc.

View on HN

I think it may be even more fun the other way, players need to rat out other players, do they have to pretend to be an AI.

It makes it easier for the AI mods to do the part and it puts the burden on the players.

Yep. "Find the AI" is broken for now, because whenever I've played it, humans can just be exceptionally rude/lewd or use super-modern slang. Making the humans try to blend in as AI is much more interesting, as a game.

A game like Press The Button, except there’s half AIs and half human players. The goal for each is to identify the others and airlock them off the ship. Constrain the tests in such a way that open chat using lewd language or whatever is impossible. I’d play that once or twice.

Where do you play that?

I like your idea of find the human. Just building on that idea a little. I know current AI detection programs don't work well. But they would be fun in the context of a game. Call it "Only Robots Allowed" and have it be a single player version of Among Us. Pretend to be a robot while trying to sabotage robot things. AI detection is applied to your conversation with other robots. And also applied to your movements. If you fail the AI detection by not emulating an AI well enough, then it's time for "kill all humans!"

This is probably the most important life skill we should be teaching in schools </pessimism>.

It would be pretty easy to beat. One sure way to tell an LLM apart from a human is to type something nonsensical and optionally repeat it multiple times. A human would inevitably answer with something along the lines of "what the hell are you on about dude???", which is something you'd never see from an LLM.

> which is something you'd never see from an LLM.

Might not see it from ChatGPT - but "never from an LLM"? Why would you think that?

So you make it learn. It collects everything humans say and matches it to the situation, replaying human conversation as needed.

With the current state, most powerful LLMs are also limited in the amount of topics they are allowed to discuss. I bet you could easily differentiate between player and AI by asking opinions on some controversial topic.

So like Battlestar Gallactica. Who is the cylon? The cylon might not even know they are not human until revealed to them later.

Reply on news.ycombinator.com