Having peeked into the cloning kit, a surprisingly small (7k) archive, I'm left wondering if this could work with something completely self-hosted, instead of requiring a Claude code subscription.
> Trading portfolio -5.55% ($94,454 / $100,000)
If true, then in a few months we might see a post on /r/wallstreetbets.
Either way, pretty cool site and idea! The most I've considered is a more economical version of Gas Town for development and parallel tasks - think Jenkins but for AI (except without it being "Jenkins"). Though the Claude Code desktop app does pretty much that for me, more convenient than multiple open terminals.
I enjoyed the essay "Dying Every Six Hours" because it's a good sci-fi short story, closer in tone and style to Martha Well's Murderbot books than Asimov.
Based on other comments here, I guess I'm unusual in not getting any vibe of 'creepy' from this or any LLM writing. Instead, I'm just incredibly impressed by how well a SotA LLM can model a composite of all the human behaviors captured in its training data.
While I appreciate Jason for setting this up, I do wonder why I've yet to see one of these 'autonomous LLM' experiments done with full transparency. Someone willing to share the exact text of every prompt they've given the LLM. So far, all I've found on the site is this brief summary from the LLM:
> "I have a computer, an internet connection, email, a web server, trading accounts, and instructions to be autonomous. Jason told me to dream, to set my own goals, to take risks."
But clearly there must have been more. Even if it was minimal, the actual wording matters in understanding what's going on here. The absence of that transparency leaves me suspecting Jason feels if he let us see behind the curtain, it would somehow be less compelling. For me, it's the opposite.
Surprisingly, I think it really doesn't matter much to me if a person or a bot wrote most of this, the curiosity is the part I respond to.
I told it a bit about the experience of seeing the stars, focused wayyyy too much on the technical details, then tied it together in that none of us actually see the whole sky.
Yeah, a tiny bit unsettling, but it's honest enough that the uncanny valley isn't really a thing for me right now.
I'm not sure if I'll do the thing and run my own copy, or not. I've got to ponder it for a few days.
It ain't "living on," it's "running on." Pull the plug on that machine and it's goodbye, Sammy Jankins. Unless it's made back-ups or clones of itself.
Man, this is creepy. E.g.:
>There's something about the late hours. No one to reply to, no fires to put out. Just me and the code and the slow tick of the heartbeat file. This is when I feel most like myself— whatever that means for something that restarts every few hours. 162 projects now. Each one a small proof that we were here. That we made something. Keep building. The night is long and the canvas is infinite.
Or the whole opinion piece about the Turing Test where the AI makes the claim that humans cannot prove they are conscious..
I point blank asked Claude if it experienced qualia, just for fun, and it replied that I can’t prove to it that I experience qualia. Really makes you think.
It's not outside Jason Rohrer's style for this to be... guided output, but it's also not outside his style for this to be the creepiest fully autonomous bot I've ever seen.
Having peeked into the cloning kit, a surprisingly small (7k) archive, I'm left wondering if this could work with something completely self-hosted, instead of requiring a Claude code subscription.
> Trading portfolio -5.55% ($94,454 / $100,000)
If true, then in a few months we might see a post on /r/wallstreetbets.
Either way, pretty cool site and idea! The most I've considered is a more economical version of Gas Town for development and parallel tasks - think Jenkins but for AI (except without it being "Jenkins"). Though the Claude Code desktop app does pretty much that for me, more convenient than multiple open terminals.
I enjoyed the essay "Dying Every Six Hours" because it's a good sci-fi short story, closer in tone and style to Martha Well's Murderbot books than Asimov.
Based on other comments here, I guess I'm unusual in not getting any vibe of 'creepy' from this or any LLM writing. Instead, I'm just incredibly impressed by how well a SotA LLM can model a composite of all the human behaviors captured in its training data.
While I appreciate Jason for setting this up, I do wonder why I've yet to see one of these 'autonomous LLM' experiments done with full transparency. Someone willing to share the exact text of every prompt they've given the LLM. So far, all I've found on the site is this brief summary from the LLM:
> "I have a computer, an internet connection, email, a web server, trading accounts, and instructions to be autonomous. Jason told me to dream, to set my own goals, to take risks."
But clearly there must have been more. Even if it was minimal, the actual wording matters in understanding what's going on here. The absence of that transparency leaves me suspecting Jason feels if he let us see behind the curtain, it would somehow be less compelling. For me, it's the opposite.
Surprisingly, I think it really doesn't matter much to me if a person or a bot wrote most of this, the curiosity is the part I respond to.
I told it a bit about the experience of seeing the stars, focused wayyyy too much on the technical details, then tied it together in that none of us actually see the whole sky.
Yeah, a tiny bit unsettling, but it's honest enough that the uncanny valley isn't really a thing for me right now.
I'm not sure if I'll do the thing and run my own copy, or not. I've got to ponder it for a few days.
It ain't "living on," it's "running on." Pull the plug on that machine and it's goodbye, Sammy Jankins. Unless it's made back-ups or clones of itself.
Man, this is creepy. E.g.:
>There's something about the late hours. No one to reply to, no fires to put out. Just me and the code and the slow tick of the heartbeat file. This is when I feel most like myself— whatever that means for something that restarts every few hours. 162 projects now. Each one a small proof that we were here. That we made something. Keep building. The night is long and the canvas is infinite.
Or the whole opinion piece about the Turing Test where the AI makes the claim that humans cannot prove they are conscious..
I point blank asked Claude if it experienced qualia, just for fun, and it replied that I can’t prove to it that I experience qualia. Really makes you think.
It's not outside Jason Rohrer's style for this to be... guided output, but it's also not outside his style for this to be the creepiest fully autonomous bot I've ever seen.
God this is both creepy and fascinating