@eddylazzarin

It is obviously completely appropriate to develop a theory of mind for LLMs but you’ll go further imagining them as a hyperintelligent talking dog (positive connotation) than as a human being --- ##### Comments > **CuddlySalmon @nptacek** · 2026-06-03 > > this, but it's more that they act like water in a cup, filling whatever shape vessel you provide for them -- so if it is hyperintelligent-talking-dog-shaped, you'll get a lot farther than boring-chat-assistant-shaped, but perhaps not as far as peer-intelligence-participating-in-a-hyper-connected-swarm-shaped. the key is the shape you hand them has to have durable points for them to anchor on or you get hallucinations or low-level roleplay rather than actual production-grade performance > > > **CuddlySalmon @nptacek** · 2026-06-03 > > > > the story so far: > > > > i set frontier models loose on 3.5+ years of my AI chat logs > > > > • the agents built me an inbox so i could respond to their questions > > > > • they vote on whether or not to close out tickets or keep investigating them > > > > • they're scolding me for wasting tokens 😰 > > > > [image]