Sequence structure notes

LessWrong sequences are not just categories or link lists. A good sequence is a guided argument spread across posts. It has a thesis, an ordering, short chapter-level claims, cross-links, and enough scaffolding that a reader can enter the author’s frame before reading every source.

What Comprises a Sequence

• A short title that names the transformation or lens.
• A one-paragraph sequence summary: what the reader will understand by the end.
• A warning / scope note: what kind of claim this is, and what it is not trying to prove.
• Three to five parts, each with a small number of posts.
• Each post has one core claim, not just a topic.
• Later posts reuse terms introduced earlier, so the sequence compounds.
• Sources are embedded as “read next” or “source packet” items, not dumped at the end.
• The sequence has a final synthesis post that says what changed in the author’s view.

Richard Ngo / Replacing Fear As Template

If someone handed over Richard Ngo’s thoughts and said “make a sequence out of Replacing Fear,” the structure would be:

1. Name the target transition.

   • From fear-based motivation to excitement-based motivation.
   • The title is not “Notes on Motivation”; it names the change the reader is supposed to undergo.

2. Split the argument into phases.

   • Part 1: identify the failure mode.
   • Part 2: build the replacement mechanism.
   • Part 3: cultivate the new mode and avoid traps.

3. Make each post a stepping stone.

   • Fear versus excitement.
   • Hidden standards inside judgment.
   • Protective strategies learned under danger and rejection.
   • Internal coercion between parts.
   • Listening to what parts want.
   • Trust through bids and boundaries.
   • Self-leadership and self-love.
   • Coercion as scarcity, trust as abundance.
   • Object-level obsession.
   • Agency begets agency.
   • Self-deception warning.
   • Determination as closing virtue.

4. Make the sequence cumulative.

   • “Fear”, “parts”, “coercion”, “trust”, “bids”, “boundaries”, and “excitement” become reusable handles.
   • Later posts do not restart; they rely on the reader carrying the handles forward.

5. End with a virtue or operating stance.

   • The sequence does not just explain a model. It leaves the reader with a way to act.

Applying This To The Sequences

The sequences should treat each major theme as a sequence:

1. Harness

   • Transformation: from “agents as chatbots” to “agents as controlled processes.”
   • Sequence shape: model plus harness; feedback loops; MLD telemetry; /goal; security and review; long-term artifact coherence.
   • Sources: Lopopolo, Codex /goal, OpenCodeMAX, George on cybernetics, Auto-Review, Marius/Watcher, Moraine.

2. Long Horizons

   • Transformation: from “just keep working” to structured long-horizon goal pursuit.
   • Sequence shape: why distant goals fail; waypoints; planning invariance; horizon generalization; triangle/quasimetric distances; product UX bridge to /goal.
   • Sources: Eysenbach / Princeton / Berkeley papers, RL for Planning, Horizon Generalization, Offline GCRL with Quasimetrics, Horizon Reduction, A Single Goal.

3. Cybernetic Economics

   • Transformation: from “sell cognition” to “own the feedback loop.”
   • Sequence shape: smart squeeze; context as dark matter; outcome pricing failure; cybernetic rollups; edge intelligence; business gravity.
   • Sources: Soren Larson, George, MentalGeorge, Harvey/Spellbook as case material.

4. Vertical Tasks / Data

   • Transformation: from “domain wrapper” to “workflow/data/eval system.”
   • Sequence shape: Harvey LAB; expert rubrics; legal AI controversy; token resale critique; what would count as a moat.
   • Sources: Harvey LAB, Harvey beef, Spellbook, WillC, Matt Ambrogi, Gabe Pereyra, PhysicianBench/LegalBench.

5. Constitutional Alignment

   • Transformation: from “rules” to “formation.”
   • Sequence shape: Model Spec; deliberative alignment; constitutions as public law/upbringing/eval target; persona formation; hidden traits; theological mirror.
   • Sources: OpenAI Model Spec, deliberative alignment, Joe Carlsmith, Owain Evans, Tim Hwang/ICMI, Goodfire/Joseph Bloom.

6. Visual Research Artifacts

   • Transformation: from “paper summary” to “navigable artifact.”
   • Sequence shape: why presentation matters; progressive visual explanation; interfaces for document dumps; sequences as research object.
   • Sources: Schmidhuber Problems visual tour, World Models, UFO browser, anti-AI-slop sequence.

Proposed Sequence Format

Each sequence page should have:

• Thesis: 3-5 sentences.
• Map: 4-8 short posts/chapters.
• Source packet: links grouped by chapter.
• Editorial notes: what claim the source updates.
• Open questions: what we do not yet believe or need to test.
• Promotion status: potential, selected, drafted, published.

This keeps the sequences from becoming a bibliography. The reader gets an argument, and the sources become evidence and texture inside that argument.