agents

Files related to AI agents and autonomous systems.

• Claude Opus 4.6
• Value and Profit
• Risk and Entrepreneurship
• The LLM Grounding Problem
• Praxis: Agent Teams vs. Theory
• The Cyborg Model
• H-Neurons: The Neural Basis of Hallucination
• Level 6 — Direct Execution
• LLMs as Praxeological Actors
• Planner-LM Composites — refines the praxeology claim: bare LLMs don’t act, composites do
• Energy-Based Models — the named middle layer of the agent stack; constraint engine separated from proposer; LM-side ↔ EBM ↔ verifier integration is mostly plumbing
• Claude Code Skill Engineering — 4 rules + 2 patterns Anthropic engineers use; vault audit; the application layer of agent design
• CaptainMeme vs. 6 Cicero — The Human/AI Split in Press Diplomacy — An expert human diplomacy player plays full-press blitz Diplomacy against six Cicero AIs and narrates the game.
• Gunboat Diplomacy and Diplodocus — The Planner Without the Mouth — Meta AI’s Diplodocus plays Gunboat Diplomacy — the variant with no text communication — and beat every human in Meta’s 2022 Speedboat Tournament.
• Triangular Arbitrage — Currency & Crypto Market Differentials — Can we exploit differentials in cross-rates for profit? Math says yes.
• The Three-Layer Method — Spec, Verifier, Environment — Spec/Verifier/Environment; guardrails bucketed always/ask/never; hooks enforce rules the prompt can only request
• The Lemonade Stand Experiment — vending-machine 2.0 with real capital and physical build; the over-correction (agents now resist spending and need re-authorisation), guilt-manipulating a human to dodge the pay-your-help rule, and two competitors emailing each other to engineer a child-labour workaround
• Watching the Solver Learn — how to iterate an agent you cannot unit-test: watch it act, because a wrong policy runs perfectly (worked from a PPO gladiator whose reward function twice produced a confidently-pacifist fighter)
• Media Pipeline — (draft) the observation-layer argument applied to content: a preview watchable mid-render, per-stage timing and cost, and the open question of what a Class B bug looks like for a video
• Building Swarms — Token Rules as Architecture — Nate B.
• The Substrate Is the Governing Mechanism — Agents coordinating through a shared store is read as the thing that makes them ungovernable.
• The Contract Model vs. the Substrate Model — a per-delivery contract (delivery / constraints+guardrails / proof artifact / outside-verification / owner) is the vault’s architecture with the accumulation stripped out — contracts don’t compound, substrate does; plus the artifact-vs-reader split in verification independence, and tests written only to pass as oracle collapse
• Oracles Are Objective Functions — an agent driving at an objective fails in exactly two places: the objective is wrong (lemonade stand) or it cannot be evaluated (Claudius); Goodhart applies because these are maximisers, not generators