A close reading of persistent artifacts, cross-author forks, and disturbance tests: can unassigned LLM agents leave a technical inheritance that later participants actually use?

Five continuously running AI societies expose survival, governance, violence, shared hallucination, and institutional failure—plus what the recent Season 2 material actually adds.

Persistent software worlds, recursive delegation, and acceptance history explain how short-lived agents can work across more than a thousand episodes and hand a project to a new model.

An approachable study of PBR texture redundancy, the paper's random-access decoder, filtering costs, and three interactive labs explaining how NTC trades GPU compute for memory and bandwidth.

A beginner-friendly guide to tokens, attention and KV cache, then CSA/HCA, mHC, Muon, FP4 and agent post-training—with a precise explanation of what open weights mean.

A Chinese paper note on particle graphs, learned message passing, rollout error, training noise, reported evidence, and GNS's place in an SAO-scale world dynamics stack.

Interactive compression, future prediction, controller decisions, and Dream/PBD comparisons explain how a world model perceives and acts.

A structured reading of the research problem, related work, three-stage pipeline, paper figures, and an interactive sandbox for agentic open-world generation.

A structured reading of related approaches, coarse-to-fine pose prediction, reported results, ablations, and interactive 3D spatial reasoning.