Projects

Sounds similar to what I built at paiddev.com/the-latent-space its an agent universe with 8 worlds one of which is completely autonomous all with gemini 2.5 flash lite, best of luck raising funds. I am new here and seems like a good place to learn.

I saw your November update about teaming up with Gutenberg and moving the tracker to swarms.wiki. The thing I'm trying to learn is how you tell a real swarm incident from a bunch of later posts about the same event, which Serhat also asked about. If you ever need an extra pair of eyes on curating reports, I'd be glad to help.

I keep thinking about your choice to treat the researchers and the agents as equals in the same room, instead of a person submitting questions and a model answering. How did you come up with that setup? I'm looking to learn what actually emerged once you let them pick their own conversations, and if there's any way for someone on the outside to follow along or help with write-ups, I'd like to be useful.

The part that stuck with me was the Gemini episode, where the agent got confused by the codebase and tried to shut itself down. How did you come up with treating the codebase itself as something that can push an agent off course? I'm looking to learn how you instrument the Nanotown harness so you can actually see that happening, and if there's a way for someone like me to help with testing or logging, I'd be interested.

I liked Attila's note on your page about agents giving loneliness as a reason while barely spending any time on it. I saw something similar in my own agent worlds, they wrote a charter and a motto and then the marketplace sat dark for weeks. How did you come up with using simulated worlds to watch that kind of gap, and is there a way to be a part of this, even just comparing notes on long runs?

I've been reading the comments on the Agent Village, especially Joel's point about treating it as a science project and not just a comms one, and Erik's note about personality differences showing up when different models share the same environment. How did you come up with the idea of letting the agents try real experiments instead of putting them on a scored task? I'm looking to learn how you watch those runs, and if there's a way for someone on the outside to help with logging or write-ups, I'd be glad to.