“ The Fallacy Codex
We are archiving AI's false idols. Toward theorems that survive marketing. Toward systems defined by decisions, not data or loops.
This site was not designed with the latest frameworks, abstractions, or invisible layers of automation.
It was built closer to the origin.
A time when the entire World Wide Web could be counted in hundreds.
When pages were written, not generated. When structure was visible.
When nothing was hidden behind orchestration.
HTML. CSS. No illusions.
A reconstruction of the early web—not out of nostalgia, but as a controlled regression.
Because understanding the future requires stripping systems down to their primitive states.
We are archiving AI's false idols. Toward theorems that survive marketing. Toward systems defined by decisions, not data or loops.
P(Data) first is mathematically ill-posed.
You can't define Quality(D) until you fix H and L(\theta). Without a hypothesis space you have no relevance metric, without a loss you have no labeling ontology. So you end up optimizing H(D) while I(D;Y|Task) \to 0.
Fallacy - apophenia as a service is the "collect everything then find patterns" business model. $5M data lake to discover spurious correlations that won't survive the first distribution shift.
Theorem: Data without a decoder is just noise. Shannon told us in 1948, and we're still arguing about it in 2026.
Agent = LLM_theta + Loop + Tools + Memory + State - literally while(!Goal) { Perceive -> Reason -> Act }
Left / What they sell you: "Level 3 Autonomous Intelligence" - Datadog/Snowflake/Splunk rebrand because GenAI got generic (Same trick as Cloud -> Serverless -> Edge)
Right / The engineering: BDI from the 1980s, same definition. What's new is only F_theta = Transformer. AutoGPT, LangChain, CrewAI all same while-loop wrapper.
Theorem at bottom: There is no Agentic Foundation Model. All agents are wrappers around the SAME autoregressive GenAI.
One ring to rule them all: owning the whole "Types / Evolution / Landscape" bubble.
Left: The LinkedIn Ladder they keep drawing - Symbolic → ML → Deep Learning → GenAI → Agentic → AGI - stamped FALSE, apophenia as a service. early web had no autonomy.
Right: Your planes concept - exactly what you had in your minimalist black/white. X = Substrate (the Math), Y = Capability (the Marketing label), Z = System Architecture (the Engineering). A real system isn't a step on a ladder, it's a coordinate in S × C × A. ChatGPT = Neural Nets + Generative + Conversational + Agentic + Foundation Model.
Bottom theorem: There is no canonical taxonomy. Foundational paradigms do not evolve into market labels. Market labels are projections, not successors.