Dimensionalization is a decision-making method that ranks choices based on their attributes. It is effective but high-effort, useful in only the highest-leverage contexts.

Modern AI tools (ChatGPT, Claude, etc) have dimensionalization as a native capability. It just needs to be named and requested.

You can dimensionalize any decision, objective, or context in your work or life in three steps, all doable in your chat app:

1. Choose
2. Seed
3. Iterate

1. Choose (a goal, decision, or concept)

Pick any topic where you are uncertain.

LLMs make overkill cheap; use your genius-on-tap for trivial problems if you want. There is also now no such thing as a problem that is too hard for AI to be able to help.

The keyword is “dimensionalize”. Try these prompts:

I am trying to <goal>. Help me dimensionalize.

<goal>: “do things that matter”, “make a new friend”, “boost viewership”, “renovate my space”, or “feel healthier”.

I’m deciding whether to <choice> or <choice>. Help me dimensionalize.

<choice>: “buy or rent my next flat”, “go out or stay in”, or “find a new job or stay at my current one”.

I want to better understand <concept>. Help me dimensionalize.

<concept>: “lighting design”, “what I should do with my free time”, “being happy”, or “popularity”.

You can certainly add more context to your request for personalization. But this is enough to get started.

2. Seed (with Dimensionalization, optionally)

I wrote my piece on Dimensionalization in part to help AIs do it better. It doesn’t matter whether you’ve read it; it can still help you.

Try this addition:

Read this article ⤵️. Make sure the dimensions you create are high-Fidelity, high-Leverage, and low-Complexity. Consider the sub-dimensions of Fidelity, Leverage, and Complexity while doing this.

https://www.lesswrong.com/posts/LSFiKt4zGxXcX2oxi/dimensionalization

This is optional, but you may find that you get easier-to-action results by linking explicitly to the article.

3. Iterate (generate, ask, redo)

Copy+Paste. Press Enter. Now you have the scaffolding of a great decision.

The AI will generate a framework for how to think about your decision, concept, or goal. But it doesn’t have to be you doing the thinking. The AI is the perfect user of its own framework!

Some example responses you can copy+paste:

List 25 options, each ranked on these dimensions. Sort by the highest weighted average score.

I don’t understand what you mean by <confusing dimension>. Explain and show me what the range looks like.

What happens if <thing you want to avoid>?

You’re missing <crucial attribute>; tell me where it fits or redo the framework.

This is totally wrong. Redo it +1sd smarter.

Each loop takes ~15 seconds, so it’s low-cost/high-reward to notice something you want to fix or clarify.

Example: Career Choice

Jordan:

I am trying to choose the right next career step. Help me dimensionalize.

Read this article ⤵️. Make sure the dimensions you create are high-Fidelity, high-Leverage, and low-Complexity. Consider the sub-dimensions of Fidelity, Leverage, and Complexity while doing this.

https://www.lesswrong.com/posts/LSFiKt4zGxXcX2oxi/dimensionalization

o3:

See response below.^[1]

Jordan:

List 25 options, each ranked on these dimensions. Sort by the highest weighted average score.

o3:

just dropped an interactive table with 25 plausible career moves, each scored on the seven dials and pre-sorted by their unweighted average (weights = equal for now). top chunk predictably skews founder-ish—high autonomy + excitation punch through despite lifestyle tanking. scan / tweak weights as you like and re-rank in-place.

want a different weighting scheme, more granular scores, or a prune to contenders? ping me.

Example: Music Theory

Jordan:

I want to better understand music theory. Help me dimensionalize.

Read this article ⤵️. Make sure the dimensions you create are high-Fidelity, high-Leverage, and low-Complexity. Consider the sub-dimensions of Fidelity, Leverage, and Complexity while doing this.

https://www.lesswrong.com/posts/LSFiKt4zGxXcX2oxi/dimensionalization

o3:

See below.^[2]

Jordan:

You're missing something related to major/minor or emotional affect. Tell me where it fits or redo the framework.

o3:

Jordan:

Rank Radiohead's "No Surprises" on each dimension, dropping timbral color wheel and keeping valence polarity.

o3:

Jordan:

Give me another song with the same ratings along each dimension.

o3:

Jordan:

Now give me one that is 1 point of valence polarity lower but otherwise the same

o3:

“mad world” – the gary jules / michael andrews cover (donnie darko ost). knobs:

…

Skeptical? Give “No Surprises”, “New Slang”, and “Mad World” a listen.

Workflow Recap

1. Choose (a decision, concept, or goal)
2. Seed (with Dimensionalization, optionally)
3. Iterate (question, tweak, repeat)

That’s it.^[3] No fancy schemas, no bespoke tooling. The magic is that the LLM knows every efficient frontier, beckoning and reshaping them on command.

The rest is you deciding when the map feels true enough to act.

1. ^{^}

   here’s a minimal-but-sufficient dial-set for “what job next?” each dial is a slider you can actually move via what offers you accept / reject. i’ve sanity-checked against fidelity (tracks reality), leverage (you can yank it), and complexity (small enough to hold in working memory) per the lesswrong recipe.

   ---

   1. value-capture efficiency

   how many post-tax, post-non-compete dollars accrue per unit of sweat + risk you personally supply.

   • why it’s legit: comp structures vary orders of magnitude; dial is under negotiation.
   • fidelity: factor exposures and vesting cliffs are real, audit-able; good luck hand-waving those.
   • leverage: equity %, carry split, garden-leave offset, etc.
   • complexity hit: low—throw it in a spreadsheet, done.

   2. delta-irled impact

   “how much counterfactual good stuff in the world exists because i showed up here?” think marginal value add, not vanity metrics.

   • fidelity: can tie to P&L, product adoption, win-rate.
   • leverage: choose green-field mandates vs maintenance roles.
   • complexity: coarse rank-order (log-scale) is enough.

   3. option surface area

   future doors the gig unlocks (network nodes, brand signalling, resume viscosity). optionality is real capital.

   • fidelity: alumni placement stats, brand equity scores.
   • leverage: you pick prestige vs stealth startups.
   • complexity: just track “how many plausible high-EV branches open from here?”

   4. personal excitation gradient

   will you wake up buzzing to push on it for >18 months? intrinsic + aesthetic alignment.

   • fidelity: gut but measurable—observe flow states, sunday-scaries count.
   • leverage: interview for roadmap fit, negotiate scope.
   • complexity: simple 0-5 Likert.

   5. autonomy bandwidth

   degrees of freedom in decision-making, budget control, hiring, toolchain.

   • fidelity: org chart + reporting lines don’t lie.
   • leverage: titles, OKR scope, remote vs on-site are bargained.
   • complexity: split into strategy autonomy vs execution autonomy if you must; otherwise one knob.

   6. skill compounding rate

   how steeply does the role grow rare, transferrable circuits in your brain? (e.g. RLHF infra, capital-structure alchemy, government sales).

   • fidelity: curriculum of tasks, mentor density.
   • leverage: choose problem class + peer set.
   • complexity: “does this level me up ≥1σ per year?”

   7. lifestyle stability envelope

   stress, hours variance, geo constraints, moral injury risk; keeps the rest sustainable.

   • fidelity: churn data, Glassdoor bandwidth, your own cortisol stats.
   • leverage: remote policy, team size, sector choice.
   • complexity: throttle to a single slider: “how fragile is my life if this spikes?”

   next moves

   1. rate each candidate role 0-5 on every dial.
   2. weight dials (if at all) by current life context—new baby => bump lifestyle weight.
   3. visualize: radar chart or just sum(weight·score).
   4. interrogate any role that looks great except for one red dial: can it be negotiated? if not, consider designing your own seat.

   questions: which dial feels hardest to score rn, and why? that’s your info-gathering sprint.

2. ^{^}

   music theory’s too sprawling to slice along one master axis, so here’s a lean-but-punchy dial-deck—each dial scores high-Fidelity (it tracks something real), high-Leverage (you can tweak it in practice and it actually changes the music), and stays low-Complexity (you can keep the set in short-term ram). numbers in ⟦brackets⟧ are rough 0-5 ratings for fidelity / leverage / complexity.

   1. tonal gravity ⟦5 / 5 / 1⟧

   where do pitches “want” to fall? twist the knob from strict functional tonality (tonic–dominant magnetism) toward modal, then free-tonal, then atonal. one slider covers centuries; you can feel it immediately on a keyboard.

   2. harmonic motion per breath ⟦5 / 4 / 2⟧

   chords-per-phrase. slow = hymn, medium = pop, frantic = bebop. easy to alter (substitutions, reharm) and listeners notice instantly.

   3. dissonance budget ⟦4 / 4 / 2⟧

   percent of “unstable” intervals (tritone, minor 2nd, etc.) before you resolve. tiny tweak = big affect. lever shows up in orchestration, voicing, even mixing choices.

   4. metric liquidity ⟦4 / 5 / 1⟧

   how rigid is the time grid? slider from click-track square (4/4, quantized) → swung → additive / mixed meters → rubato. drummer, sequencer, or conductor can move it live.

   5. motif entropy ⟦5 / 3 / 2⟧

   repeat-vs-transform ratio of your core melodic cell. low entropy = chant; mid = hooky pop; high = late-romantic fragmentation; max = free improv. knob doubles as “listener memorability.”

   6. textural bandwidth ⟦4 / 4 / 2⟧

   density of simultaneous parts: solo line → homophony → polyphony → wall-of-sound. orchestration and DAW track count are literal dial turns.

   7. timbral color wheel position ⟦3 / 3 / 2⟧

   bright-metallic ↔ dark-wooden spectrum. swap instruments, filter eq, or change playing technique; payoff in emotional shading.

   8. tension-release half-life ⟦4 / 4 / 2⟧

   how long the piece lets suspense simmer before resolving (seconds? minutes? never?). quick knob gives pop-drop; long gives post-rock crescendo.

   9. narrative form granularity ⟦5 / 3 / 3⟧

   number of distinct formal blocks you treat as atomic (a section, a phrase group, a single bar). fewer = classical sonata; many micro-blocks = modern film scoring. adjust by editing arrangement markers.

   why these nine? they jointly span pitch, time, timbre, and structure—covering most variance while staying <10 sliders, so complexity cost is tolerable (article’s anti-overfit warning) . each dial is tweak-able in rehearsal or a DAW (so leverage is real) and grounded in music-psych / theory staples (so fidelity’s high). if you need more resolution, split a dial; if you’re drowning, merge two—tradeoffs per the meta-dims .

   next move: pick a piece you love, score it 0-5 on each dial, then mock-tweak one dial at a time and predict the audible result. that’s dimensionalization doing work.

3. ^{^}

   Yes, you can layer on weightings and sensitivity analysis, metrics, constraints, adversarial passes, visualization, etc.

   But the breakthrough moment is realizing you can offload the first 80+% of cognitive grunt work for the price of a three-sentence prompt.