Dasein Manifesto

Commissioned illustration: human brain cognitive modules — visual cortex, hippocampus, TPJ, mPFC, language areas, motor cortex — transitioning into AI specialized subsystems. The only existence proof of general intelligence we have is the human brain, and the brain is not one general reasoner with a long context window. It's a coordinated society of fast specialized systems — vision, language, theory of mind, memory, motor — running in parallel under a slower general substrate. Cognitive science figured this out four decades ago. AGI engineering is about to. LeCun's world-models argument and Karpathy's LLM-OS framing point at this same shape from different angles. We're building one of the organs. The five theses. One: Intelligence is not one general reasoner. It is a society of fast specialized subsystems coordinated by a slower one. Two: AGI will have the same architecture. Bigger LLMs are necessary, not sufficient — world models, episodic memory, and persistent state are the missing organs. Three: Memory is the canonical specialized subsystem — fast hippocampal pattern completion in milliseconds, slow cortical consolidation in hours. Four: Dasein is the hippocampus for LLMs. Four hops in 1,057 milliseconds, fully cited. Five: Every serious AGI in five years has one. We intend to be it. The five theses laid out as a single quotable visual designed for X, LinkedIn, and Hacker News screenshot sharing — numbers and benchmark figures live in the footnote receipts, not on the visual itself, so the diagram propagates without dating. What's next: We have the receipts on the four pillars, a roadmap for the next three subsystems we're building — context optimizer, meta-learning, self-improving context layer — and the capital to ship them. We don't have everything an AGI needs. Nobody does. We have the layer we know how to build, and we're shipping it. Build with us if you ship a product where retrieval quality decides whether the agent reasons. Talk to us if you're a researcher building one of the other specialized subsystems and want to compose. Watch us if you want to see if this thesis holds. — Nick & Surya. Footnote 1: Fodor, The Modularity of Mind (1983). Kanwisher on functional localization, including FFA. Saxe on TPJ and theory of mind. Kahneman, Thinking, Fast and Slow (2011), for the System 1 and System 2 framing. Footnote 2: McClelland, McNaughton, O'Reilly, Why there are complementary learning systems in the hippocampus and neocortex (1995). Footnote 3: Each pillar beats the published state of the art on its own benchmark. Compression vs Google's TurboQuant — 27/27 byte-matched wins. Per-query α — 0.40 ms, +12 percentage points R@1 on FEVER. Per-query top-K — 68% fewer tokens, R@1 preserved. Sub-second 4-hop agentic search — 1,057 ms instrumented timing.