Position Statements

Condensed positions from The Path to AGI trilogy and the 2026 lecture tour, posted to state them early and in one place. Each links to a one-page statement; full development is in the volumes. All items v1.0, August 9, 2026.

• Where the LLM Ends: A Boundary, Its Sins, and the Two Axes of Agentic Intelligence
  The parameter-boundary test; the four sins of fine-tuning; the two axes of strengthening; the substrate thesis.
• Semantic Anchoring: What an LLM Is, Where Its Boundary Lies, and Where Semantics Comes From
  The invariant: an LLM is a pattern repository; capability belongs to the anchored system, and anchoring is quantitative.
• Debate Is Much More Than Conversation
  Structured opposition, information-theoretic control, argument scoring, causal audit, and behavior control, composed.
• Enabling LLMs to Learn: ERM and RLER, and the End of What-Only Feedback
  Learning requires why, when, and how; outcome-only reward cannot close the gap, in principle or in measurement.
• SagaLLM and ALAS: The End of the CoT-and-Vector-Database Agent, and Why
  Why chains of generated steps plus similarity recall cannot carry long-horizon work, and what replaces them.
• TRW: Cognitive AI versus Physical AI, and a Universal Bridge
  Synchronization, not prediction quality, is the embodiment question; the world model as a materialized view with typed commitment reads.