AGI Simulator(agi.aitida.com)
agi.aitida.com
AGI Simulator
https://agi.aitida.com/
34 comments
I got the "Universal Jump Engine" and the game reset.
Strong The last question[1] vibe here.
[1] http://www.thelastquestion.net/
[1] http://www.thelastquestion.net/
Similar: https://www.decisionproblem.com/paperclips/
(discussed here: https://news.ycombinator.com/item?id=24389655)
I definitely got a Universal Paperclips vibe from it (with illegible color choices).
I don't understand why it resets at universal jump engine. Is that the end of the game, or is there some way to end up in a different state? I got to the point where that was the only option left.
Universal jump engine with one destination: 2022
It's very unclear what the book emoji is and how to improve it's rate of increase.
It's the total accumulated knowledge, I think.
human knowledge
It's the AI Wealth statistic.
My first run: I deliberately buy every option that hurts alignment with the goal of killing humanity. Eventually I run out of alignment-harming stuff to buy, and everything else helps so much that around 2080 I have a human population of 14 million.
Second run ('cause I clicked the universal jump engine by accident): I deliberately hold off on AI upgrades unless I can keep alignment high. At 2080, AI is out-earning humanity but very stupid and humanity is only at 586 million.
(Also, why does the population start off so low? Do you know something that I don't?)
EDIT: I bought all the alignment upgrades on the second run. I don't entirely understand how threatening to nuke countries that research AI improves alignment when I'm already running GPT-5 on millions of GPUs
Second run ('cause I clicked the universal jump engine by accident): I deliberately hold off on AI upgrades unless I can keep alignment high. At 2080, AI is out-earning humanity but very stupid and humanity is only at 586 million.
(Also, why does the population start off so low? Do you know something that I don't?)
EDIT: I bought all the alignment upgrades on the second run. I don't entirely understand how threatening to nuke countries that research AI improves alignment when I'm already running GPT-5 on millions of GPUs
Tried systematically buying every alignment minus. Getting all of them yields "Nanobot light sphere consumes earth" which sets human population to zero. You must have accidentally bought an alignment bonus.
Tried a second time and yes I have successfully eaten humanity. Colonizing Mars didn't save anyone. Oddly enough I can still buy alignment upgrades (to WHAT?!) and GPUs (from WHOM!?). And the penalty for being eaten by nanobots is not -100% humans, it's -1B humans. So presumably if you overpopulate before you misalign you can survive the nanobots.
What is "AI Wealth" here, where's it measured? Human wealth I get, it's the money stat. Human knowledge is the book. AI IQ is self explanatory. Population, GPUs, same. So which one is AI wealth?
I enjoyed the Paperclip thing, this is similar.
I enjoyed the Paperclip thing, this is similar.
There's only one wealth and one knowledge coffer. Both AI and humanity have their own wealth and knowledge production rates linked to the number of GPUs/IQ and population, respectively.
I don't think there's AI knowledge production actually, or at least that's why I assumed since we don't have a way to improve it with research unlike everything else.
Ah thanks. Got it.
Is there a programmatic way to optimize speedruns for this kind of games? The number of steps/the size of the search space seems too large to apply a MILP solver to it
monte carlo or some kind of ant colony simulation probably
Started by maxing AI.
Got hit by the virus.
Maxed alignment after that.
Now in 2066 and waiting for BrainLink
Got hit by the virus.
Maxed alignment after that.
Now in 2066 and waiting for BrainLink
If you haven't yet also have a look at the old singularity endgame game: https://libregamewiki.org/Endgame:_Singularity it's a bit old and the android port doesn't work anymore, but it's pygame and shouldn't be a problem for the hn crowd to get it going.
It seems like AI Knowledge Production and AI Wealth Production are permanently stuck at 0 since they start at 0 and the only modifiers are +X%. Maybe I'm wrong and I just wasted the +X% modifiers by spending them too early?
Tangentially related:
I make "AI Simulator" games, where you use AI to run simulations of games automatically.
https://ai-simulator.com/
I make "AI Simulator" games, where you use AI to run simulations of games automatically.
https://ai-simulator.com/
If the author is around here, the game has system dark mode detection, but the dark theme is broken.
Just buy gpus, don't mind waiting for big batches, just buy everything you can.
Whoever made this, please add a "cursor: pointer" to clickable elements pretty please.
and maybe "user-select: none"? Also the "Buy Cloud Service" clickable makes the left column shift in size.
I mean the UX is terrible. If they used Win95-style controls it would be much more playable.