Visual Thought AGI Blueprint 424 pages (visualthoughtagi.com)

A 424‑page "Visual Thought AGI" blueprint has been released proposing a concrete architecture for human‑level general intelligence built around continuous, internally generated visual scenes. Instead of treating intelligence as language‑token prediction, the design grounds cognition in dynamic multimodal world‑models that construct and update imagined visual environments, run forward simulations, and evaluate counterfactuals. Key subsystems described include generative scene imagination, spatial reasoning, symbolic/meta‑cognitive layers for reflection, and integrated perception‑memory‑prediction pipelines—enabling planning, causal simulation, problem solving, and creative synthesis in a way meant to resemble human mental imagery and prefrontal reasoning. The blueprint matters because it targets shortcomings of current large language models—unstable mental models, poor physical causality simulation, and limited introspection—by unifying generative vision, symbolic reasoning, and self‑reflective control into an implementable AGI stack. Technically, it emphasizes continuous internal simulation (updating scenes with incoming data), hybrid neural‑symbolic components, and explicit agency modules for decision‑making. By publishing a detailed, open framework for researchers, developers and policymakers, the project signals a shift toward multimodal, simulation‑based AGI research and surfaces concrete safety, alignment and governance priorities for systems that more closely mirror human thought.