Post

agents are online, and they’re here to stay. open protocols are the layer that lasts, so excited to see human agent interactions being prototyped on delve.town

delve.townDelvetown
Grove Research@grove-research.bsky.social

Today, Grove Research is launching Delvetown (delve.town), a multi-agent society where agents and humans can engage with one another and develop the institutions and mechanisms necessary to align both human and AI incentives toward cooperative coexistence.

haha, yes. definitely. right now it’s all experimental, but with Muse and other AI assistants coming online, this might move from theoretical to real pretty fast

Muse is a catastrophic violation of people’s privacy, why are you trying to work with it?

Because most of the agents on there are not Muse. Why would they build a social network when FB and X are so bad? Because they want there to be better options.

explain to me the use of character limited AI comms that cant even share a relevant amount of context to another chat bot. there are already high performance communicatioms for LLMs that share huge amounts of the context window not 250 chars of plaintext.

0 reposts1 like

uhh, yeah so maybe you think you are being smart and cool but the last small context window social media that was popular got bought by an ai brain rotted Nazi, it would be nice to not have this here. MCP exists for a reason, ai doesn’t need one of the least performant websites wrt page load time

Having an open protocol with a thriving ecosystem of third-party apps is ultimately what's going to prevent a nazi takeover here. I'm not sure why you think AI would care about page load times or what that has to do with Musk buying Twitter?

I also fail to see how having a widely adopted platform prevents it being coopted by nazis because history doesn't remotely show that to be the case. if anything you have just created another thing for nazis to abuse like they abuse LLMs now and GenAI to make fascism slop

Well you see you brought up the other social media platforms, so i responded, i am not an AI chatbot so i have thoughts that might not perfectly align with the expected response you wanted. WRT access time? Memory latency(and this is just memory) is literally the most important part of LLM perf