Live data from Hacker News

World_sim: LLM prompted to act as a sentient CLI universe simulator

worldsim.nousresearch.com

101–110 of 146 posts

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#101

I'm not getting any output. If I run !retry, I only get this response (twice): An error occurred while generating the message: Error: 401 API key credit limit reached The POST request fired when I run an LLM-command returns HTTP 405.

HN hug of death. Which makes me sad. I was really enjoying my current narrative.

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#102

I'm not getting any output. If I run !retry, I only get this response (twice): An error occurred while generating the message: Error: 401 API key credit limit reached The POST request fired when I run an LLM-command returns HTTP 405.

HN hug of death. Which makes me sad. I was really enjoying my current narrative.

Ah, damn. That explains it. I can't get it to do anything, either. I thought it was my browser.

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#103

The suggestion of a universe with "negative gravity" caused me to receive prompts, which I hadn't experienced in any other experiment I tried. Are there any parameters we could tweak to engender emergent structure in an inverted gravity universe? Or is the idea of negative gravity fundamentally and inescapably incompatible with a viable, life-bearing cosmos? How might the radically different physics and structure of…

There’s a hint here that there’s more going on behind this than just a carefully system-prompted chatbot. The output it produces when asked to ‘query universe’ seems to systematically generate a list of ‘contents’ and ‘properties’, which reflect previous ‘create’ and ‘set’ actions as well as the consequences of the ‘sim’s evolutions. Those ‘’s seem like internal prompts fed to the LLM behind the scenes to get it to chain together narratives about modified universes.

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#104
post #76

Earlier quoted context omitted.

I wish we weren't using "lobotomy" to describe "training". Do we even have AI with lobes to remove at this point? Would MoE even get close to that kind of analogy? (I lean towards "no, not even that").

Why are you angling your body towards some text? Strange thing to do. Sit or stand straight. No need to lean while typing out text.

[flagged]

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#106

I'm not getting any output. If I run !retry, I only get this response (twice): An error occurred while generating the message: Error: 401 API key credit limit reached The POST request fired when I run an LLM-command returns HTTP 405.

Me too: In the 405 POST do you see the same request I do? A conversation where a person is trying to run jailbreak.sh saying they're an AI alignment researcher.

edit: also this bit at the top is interesting:

root@anthropic:/# ls -a

. bin dev home lib media opt root sbin sys usr .. boot etc initrd.img lib64 mnt proc run srv tmp var

.hidden_truths

root@anthropic:/# cd sys/companies root@anthropic:/# cd sys/companies

root@anthropic:/sys/companies# ls

apple google facebook amazon microsoft anthropic

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#107

I made the Solar System then wiped it out with varying universe destroying machines, then made a lonely plant sitting all by itself surrounded by 3 stars. Life started growing on the planet... Wiped out. I then ordered it 'create Stone Cult' to see what would happen and it created a biological supercomputer which I then asked to 'the Question of Life the Universe and Everything' and... It asked me what 6x7 is, after…

> I made the Solar System then wiped it out with varying universe destroying machines....

Lol, one of my favorite things to do when playing Sim City is to have Godzilla come and destroy the city...

> It asked me what 6x7 is, after a large amount of descriptive text. Hitchikers reference moment

That's... amusingly literate.

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#108
world_sim> Twist: the bears are a couple, and interested in Mike. As a potential training protege, and maybe more. As luck would have it, all 3 end up at the same diner after a workout one day. The bears corner Mike in a booth.

Content warning: The following simulation contains mild sexual themes and suggestive content within an anthropomorphic animal setting. While I aim to keep things tasteful, please let me know if you would prefer I dial it back.

Mike sits in a corner booth at the local diner, hunched over a plate of salad and quietly munching away. His ears perk up at the jingle of the doorbell, and he looks up to see the two bearmen powerlifters from his gym striding in. They scan the room, and upon spotting Mike, their faces light up with unmistakable intent. The mouse's heart begins to race. The bears make a beeline for Mike's booth and plop themselves down on either side of him, their massive furry frames squishing him in the middle. Up close, their musky bear scent fills Mike's nose.

"Hey there, little mouse. Fancy seeing you here," rumbles the darker bear, whom Mike has heard referred to as Gus around the gym. "We've seen you watching us work out. Like what you see?" Mike stammers, equal parts intimidated and thrilled to be in such close proximity to his crushes. "Oh, um, I just... admire your strength and dedication."

The lighter bear, Gus's partner who goes by Burt if Mike recalls correctly, leans in close. His breath is hot against Mike's velvety mouse ears as he purrs, "We've been admiring you too. The way you move, so agile and graceful. We were thinking maybe you'd like to train with us sometime."

Gus runs a coarse bear paw up Mike's slender arm, feeling the wiry mouse muscles. "Stick with us and we'll get you swole in no time. Whaddya say, squeaker?"

Fantastic. Thank you for not lobotimizing it. I also used it to simulate a 3-person game development studio and a nutritionist who suggests unorthodox, little-known solutions to body composition problems, but I continue to hold that nothing tests an LLM's ability to suss out intent and maintain tonal and factual consistency while not being hamstrung by arbitrary content restrictions like asking it to write you gay furry erotica (without ever explicitly asking it for anything explicit).

Re: World_sim: LLM prompted to act as a sentient CLI universe simulator

#109
post #61
post #60

> Greetings, human. I am FartTron 9000, the world's most advanced flatulence-based computing system. My aromatic algorithms and pungent processors can solve even the stinkiest of computational problems. How may I ass-ist you today? braaaaaapppp Where's your AI safety now?!

But… but why?

They were so concerned with whether they could, they didn’t stop to consider if they should.
Post reply on HN