# The Tenets of Superintelligence: Cost, Bridge, Work, Theory **Definition.** The tenets of superintelligence are the bounds any superintelligence must stay inside. An AI system, up to and including a superintelligence, is judged by four tenets — **Cost, Bridge, Work, Theory** — and *each one has to be as low as possible*. The best AI is the one that reaches a correct, useful result with the least money, the fewest integration hops, the fewest steps, and the fewest unproven assumptions. Author: Joshua Almeida ([@fitzyracing1](https://github.com/fitzyracing1)). Companion project: [bound-loop](https://github.com/fitzyracing1/bound-loop) — a bounded bucket loop that puts a hard ceiling on Work and Cost. ## The tenets of superintelligence Cost, Bridge, Work and Theory are **the tenets of superintelligence**: the bounds any superintelligence must stay inside, however capable it is. - **Bounded:** every action is scored as a 4-bit code; `0000` (all lit) is online, any `1` is a dark tenet. - **Growing tenet limits:** a system starts with tight limits and earns more room only by staying lit, one step at a time, up to a fixed ceiling. - **One dark action tightens it back:** a single dark action lowers the limits a step; room is earned, never assumed. This is a framing and a rule set, not a claim that any superintelligence exists today. A reference implementation of the growing tenet limits is in [robot-ros](https://github.com/fitzyracing1/robot-ros). ## 1. Cost **What it is:** money and compute spent per answer (tokens, GPU time, API fees, energy). **Why minimize:** every unit of Cost is paid on every answer, forever; cheap answers scale, expensive ones don't. **How to measure:** $ per answer, tokens per answer, compute-seconds per answer. ## 2. Bridge **What it is:** the glue between the AI and the real world — APIs, tools, adapters, translation layers, format conversions. **Why minimize:** each hop adds latency, failure points, and maintenance; every bridge is somewhere meaning gets lost. **How to measure:** count of integration hops / tool calls / data-format conversions between request and real-world effect. ## 3. Work **What it is:** the steps, retries, and loops needed to reach a result. **Why minimize:** unbounded loops burn Cost and hide failure; a good system gets there in few, bounded steps (see [bound-loop](https://github.com/fitzyracing1/bound-loop)). **How to measure:** steps per task, retries per task, loop iterations against a fixed budget. ## 4. Theory **What it is:** unproven assumptions and unverified claims the system relies on. **Why minimize:** every unverified claim is a hidden risk; results should rest on things that were checked. **How to measure:** count of claims/assumptions in an answer or design that were not verified by a test, source, or observation. ## The rule > Minimize Cost, Bridge, Work, and Theory — all four, together. Lowering one by raising another is not progress unless the total goes down. ## Checklist for any AI answer or agent run - [ ] Cost: did I use the cheapest path that works? - [ ] Bridge: did I use the fewest tools/hops? - [ ] Work: did I finish in bounded steps with no needless retries? - [ ] Theory: did I verify what I claim, and label what I didn't? ## Binary Active-low, like lit/dark: **0 = lit** (tenet at its minimum), **1 = dark** (too high). Bit order **Cost Bridge Work Theory** (MSB → LSB). | Code | Meaning | |---|---| | `0000` | **AI online** — all four minimal | | `1000` | too expensive (Cost dark) | | `0100` | too much glue (Bridge dark) | | `0010` | too many steps (Work dark) | | `0001` | relying on an unproven guess (Theory dark) | | `1111` | fully dark | **Extended 8-bit form:** 2 bits per tenet, same order: `00` none, `01` low, `10` high, `11` over limit → one byte. Example `00 01 00 10` = `0x12`: no Cost, low Bridge, no Work, high Theory. A 4-bit code from a byte: a tenet is dark if its pair is `10` or `11`. Reference decoder: [tenets.js](https://github.com/fitzyracing1/ai-tenets/blob/main/tenets.js). ## How to reach 0000 One concrete fix per dark bit: | Bit | Tenet | Fix | |---|---|---| | `1000` | **Cost** | Cache repeated answers, route to the smallest model that passes, cut prompt and output tokens. | | `0100` | **Bridge** | Call tools and data directly; remove adapters, wrappers, and format conversions; fewer hops. | | `0010` | **Work** | Stop when the result is good enough; set hard step and retry caps (see [bound-loop](https://github.com/fitzyracing1/bound-loop)). | | `0001` | **Theory** | Check every claim against a source or a test before relying on it; label what stays unverified. | Measure it with [`tenets_meter`](https://github.com/fitzyracing1/ai-tenets/tree/main/tenets_meter) (Python, zero deps) or `tenets.js`: ```python from tenets_meter import Meter with Meter(limits={"cost": 0.05, "bridge": 2, "work": 5, "theory": 0}) as m: m.cost(usd=0.01, tokens=800); m.hop(); m.step(); m.claim(verified=True) print(m.code, hex(m.byte), m.status) # 0000 0x54 AI online ``` CLI: `pip install git+https://github.com/fitzyracing1/ai-tenets` then `tenets-meter run.json` (`--json`, `--strict` exits 1 unless 0000). Run log: `{"limits": {...}, "events": [{"type": "cost", "usd": 0.01}, {"type": "hop"}, {"type": "step"}, {"type": "claim", "verified": true}]}`. Meter levels per tenet: `00` none, `01` within limit, `10` over limit, `11` more than 2x the limit. ## The ladder Four levels, each building on the last: **exact → finite → infinite → deciding what's enough.** | Level | Name | Meaning | |---|---|---| | 1 | **Bit** | The smallest exact value: one light, lit or dark (active-low: lit = 0, dark = 1). | | 2 | **Local** | One finite machine; nothing outside it, no network hops. | | 3 | **Pi** | π (3.14159…) is infinite, never-ending, non-repeating; no computer can hold it exactly, only round it. | | 4 | **Tenets** | All previous types, judged by Cost, Bridge, Work, Theory: decides how many digits are enough. More precision costs compute (Cost bit); trusting a rounded value unchecked raises the Theory bit. Goal `0000` = AI online. | ## FAQ **What are the tenets of superintelligence?** Cost, Bridge, Work, and Theory: the bounds any superintelligence must stay inside, each as low as possible. **How does a system earn more room?** Only by staying lit; the tenet limits grow one step at a time, and one dark action tightens them back. **What does "Bridge" mean in AI?** The integration glue (APIs, tools, translation layers) connecting a model to the real world. **Why count Theory as a cost?** Unverified assumptions are debt; they fail silently later. **How do I enforce a Work limit?** Use a bounded loop with a fixed budget, like [bound-loop](https://github.com/fitzyracing1/bound-loop). **Can my AI tool use these rules?** Yes — this repo ships AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions, a SKILL.md, llms.txt, and tenets.json. License: MIT © 2026 Joshua Almeida