Losing the Thread
Autoregressive Drift in Generative AI and What Comes Next.
First Principals: Why fluent generative systems lose objective fidelity across long outputs, agentic workflows, and model scale—and why generation-time control becomes necessary.
-
01 Article 1Objective Pursuit Is Not Objective Origin
The current AI discussion often conflates two capacities: pursuing an objective and originating one. Clarifying this difference is essential.
Read article -
02 Article 2Parsing Intelligence: From Maslow to Tinbergen
Once the distinction between objective pursuit and objective origin is established, the next problem is explanatory discipline.
Read article -
03 Article 3From Hierarchy to Control
What happens when a system begins to act over time within a structure of objectives it did not fully originate?
Read article -
04 Article 4What the Transformer Actually Changed
The transformer gave modern AI a far better engine. It did not, by that fact alone, provide a steering system.
Read article -
05 Article 5Fluent Failure: Why Capable Models Drift Smoothly
Probability, Weights, and the Logic of Local Continuation
Read article -
06 Article 6Decoding Is a Control Surface
Where Capability Becomes Behavior
Read article -
07 Article 7Drift, Named
How Local Continuation Loses the Global Objective
Read article -
08 Article 8Why Long Tasks Break More Than Short Ones
Length Is Not Just More Output
Read article -
09 Article 9The Current Toolkit and Its Limits
Prompts, Retrieval, Fine-Tuning, and Review
Read article -
10 Article 10When Agents Inherit Drift
From Generated Text to Generated Action
Read article -
11 Article 11Bigger Models, World Models, Same Problem
Scale Improves Capability. It Does Not Eliminate Drift.
Read article -
12 Article 12The Case for Runtime Control
Why Reliability Has to Happen During Generation
Read article -
13 Article 13Equilibrium-Constrained Decoding: Holding the Thread
A Runtime-Control Approach to Objective Fidelity
Read article -
14 Article 14The ε Series as a Governance Artifact
Making Objective Fidelity Observable
Read article -
15 Article 15The Economics of Objective Retention
Governable Scale and the Cost of Trust
Read article -
16 Article 16Magnifica Humanitas and the Question AI Cannot Answer for Itself
What Technology Is Made to Serve
Read article -
17 Article 17On the Regress Objection: Why ECD Moves the Alignment Problem to a Place Institutions Already Know How to Govern
Final coda to Losing the Thread.
Read article