91–100 of 121 posts
To be crass and direct, > The goal is to democratize building your own simulation environment with AI agents. Would love to see the community build more complex interactions on top of this. Let us know what you think! why? what's the point?
Irks me when people shoe-in the word "democratize".
The etymology of the word implies rule by a demos (demos / kratos), a demarcated ingroup of people such as the citizenry of a country. And yet, so many of these 'democratising' technical endeavours keep control of the project strictly behind lock and key. There's nothing democratic about it - they just want popularity. Is 'popularise' an undesirable word now? I wonder why the euphemism treadmill has had to advance in…
Could we not be living in a more advanced version of this same project?
Haha yes the most commonly used argument is (afaik) that if we could create a 'complete' virtual world, we would, and projects like this and how enthusiastic we are about it suggest that that's indeed the case. (The argument then of course continues that if we would, our world's inhabitants would some day too, ad infinitum. Given such a stack, the chance that we're actually still in the top layer is very small.)
Earlier quoted context omitted.
Haha yes the most commonly used argument is (afaik) that if we could create a 'complete' virtual world, we would, and projects like this and how enthusiastic we are about it suggest that that's indeed the case. (The argument then of course continues that if we would, our world's inhabitants would some day too, ad infinitum. Given such a stack, the chance that we're actually still in the top layer is very small.)
My counterargument to this is that people simulating our world would be simulating all the bad stuff in it, which would be cruel and unethical. I think if humans were simulating us, they'd simulate a more utopian world. And if curious-but-indifferent aliens/robots simulated us, they wouldn't be wasting compute on simulating all the boring and uninteresting things we have in our world. I just don't see why would an in…
I'm reminded of the Culture novel Surface Detail, which basically revolves around d the ethics surrounding simulating consciousness. There is a society in the book which runs afterlife simulations, including a pretty horrible hell. Conflict over whether this should be allowed to continue is one of the main drivers of the plot.
Earlier quoted context omitted.
This is made by a16z, ostensibly for business reasons. I am asking what their angle is. As far as I know, they don't have much history in making things with "no extrinsic objective" - I could be wrong though.
You misunderstand what a16z-infra is. This is built by Convex, the game engine being used.
That GitHub org is a16z.
Their other repos are also showcasing various AI toolings, often the tools are also startups they invest in.
Earlier quoted context omitted.
There is something nice in the sense that simulation appears to be completely unmonitored. If my Sims started producing scans that approximated the hardware of my Gaming PC, I would probably stop the sim... assuming I was closely monitoring.
Why?
I built something very similar when the paper came out- watch out for the instances where GPT returns both sides of the conversation (e.g. interpreting a prompt with chat history as being representative of what the output should look like, and putting out both "Jack" and "Jill"'s text). The first sample I went to suffered that problem (output from "Lucky's perspective below.) "Hey Stella! It's so great to see you! Yo…
Did you have any luck mitigating this problem?
Ie if the LLM is playing a character "Jack" and you are playing the character "James" your query might be "_only reply for your character Jack_ James: I pick up the sword and then turn towards Jack".
It can also be used to influence behaviour as LLMs often get stuck repeating (not word for word) the same events/description, ie greeting your character over and over again and not moving on, LLMs aren't great at having autonomy/agency in the flow of a conversation, I think this is best done by not providing the entire history of the conversation but instead distilling it to relevant information for the current query.
But that can also be mitigated manually by bumping their character with underscores ie "_Jack asks what James wants to order_ James: I return Jack's greeting and peruse the Tavern's menu board".
Earlier quoted context omitted.
Haha yes the most commonly used argument is (afaik) that if we could create a 'complete' virtual world, we would, and projects like this and how enthusiastic we are about it suggest that that's indeed the case. (The argument then of course continues that if we would, our world's inhabitants would some day too, ad infinitum. Given such a stack, the chance that we're actually still in the top layer is very small.)
Here's a fun little short story that touches on the idea on not being the top layer of the sim. https://qntm.org/responsibility
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 th…
I built something very similar when the paper came out- watch out for the instances where GPT returns both sides of the conversation (e.g. interpreting a prompt with chat history as being representative of what the output should look like, and putting out both "Jack" and "Jill"'s text). The first sample I went to suffered that problem (output from "Lucky's perspective below.) "Hey Stella! It's so great to see you! Yo…
I also noticed this when browsing the demo village, some conversations it ran through and did both parts.