Researcher-Operator currently on garden leave. Formerly: Two Sigma (Quant Research + Mgmt) / OnDeck (Data science in lending) / BlackRock (Bond desk quant). I hope my thinking can be helpful to you!
My Substack: https://jordanmrubin.substack.com
My LinkedIn: https://www.linkedin.com/in/jordanmrubin/
Empathy is a multi-dimensional operation and there are many ways to do it. Here are some examples of dimensions along which you can vary empathy:
How much do you try to literally “feel their feelings” vs feel your own feelings vs feel no feelings?
How charitably (or hostilely) do you try to read their situation?
How focused are you on actions they could take?
To what extent do you try to understand the story of what took them here vs focus on the present moment?
How focused are you on what is good for them vs on what is good for both?
You don’t have to do empathy the same way in every situation. But have you ever tried to max out on “feel their feelings” or “charitable read”?
Here is o3’s charitable read of nail-woman:
felt-sense override – pain is embodied confusion. before she can orient to “extract object,” she needs another nervous system to mirror how disorienting the signal is. otherwise her interoceptive map keeps screaming “unsafe,” so any fix attempt feels like an attack on the only reliable data she has left.
status & competence threat – accepting a drive-by solution telegraphs “i couldn’t notice a metal spike in my own skull.” that’s ego-annihilating. she’s fighting for narrative sovereignty, not against physics.
pattern-matching past dismissals – likely history of “just calm down” responses. the nail scene retriggers that cached pattern; she’s guarding against another instance of emotional rug-pull.
speech-act mismatch – she’s broadcasting affective data; partner keeps replying with instrumental commands. protocol violation feels like deafness. she insists “it’s not about the nail” bc rn the channel is meta-communication, not mechanics.
Let’s do one example. Suppose you are trying to get access to celebrity participants for your music videos, since we’ve identified that as high leverage. Let’s start by listing some things that could meaningfully lead to celebrities agreeing to participate:
One of the celebrity’s friends recommends that they work with you
The celebrity sees your video and wants their work treated similarly
Your agent connects with the celebrity’s agent and arranges a quid pro quo
You use archive footage of a celebrity in your video and get their permission after the fact
...
Now we can filter for what is in your control:
You can try to befriend a celebrity’s friends
You can try cold-sending your work to celebrities
You can hire an agent and ask that they help you work with celebrities
You can try to find footage of celebrities who might be amenable to this kind of fair use
...
Of these, hiring an agent strikes me as the most certain way to achieve the outcome.
If we had like 50 options (we can get that by using an LLM), then it would make sense to group them and optimize for Fidelity and Complexity. But for now we are focused solely on a few high-leverage actions and I’m not using an LLM, so I won’t go through the whole dimensionalization process.
If you again find yourself thinking “but I can’t befriend a celebrity’s friends” or something like that, you can just do the exact same exercise to dimensionalize that outcome.
Thanks. Here’s what I would suggest if you want to do Dimensionalization by hand starting with Leverage:
Start with actions you can take, and filter for those that would result in things that matter to your outcome. Or,
Work backwards from things that would matter to your outcome, and filter for those that have triggers that are at least partially in your control.
Then, once you’ve made a list of atomic actions with high Leverage, you can group them in various ways to reduce Complexity and align to your model of ‘how the outcome works’ to increase Fidelity.
FWIW, it feels obvious to me that having celebrities participate in your videos lends credibility (strong social proof) that could lead to additional commissions. Fine if your target commission audience would be put off by this; in that case, you should find audience-relevant signals of credibility. But credibility is probably going to be important regardless of how you achieve it.
Hi—I love the question.
First, it matters what you choose to dimensionalize: “Dimensionalize Posting Instagram Content” will give you different results from “Dimensionalize Soliciting More Commissions”. And “Dimensionalize Posting Instagram Content with respect to its impact on Soliciting More Commissions” will give you still different results. The key is that dimensionalization tries to give you levers to affect a specific decision or outcome.
Second, I strongly recommend using LLMs to help you dimensionalize. Without domain expertise in these areas, I am not going to provide much edge vs an LLM. You can take your comment, paste it into an LLM, link to this post, and ask for dimensionalization of whatever you want using this framework. Part of the goal of the post is to make it easy for me (and others) to do this :)
Here’s an example, straight from o3:
| axis | 0 ←—— dial ——→ 5 | why it matters |
|---|---|---|
| authority signal | personal art dump → visibly paid, tagged celebs, bts w/ shotlist overlay | commissions need trust; ig ranks “creator credibility” too |
| genre fit | random aesthetic soup → laser‑targeted to niches you want | clients self‑identify; boosts save+share rate, a heavy reel weight |
| commissionability clarity | no hint you’re for hire → explicit cta (“dm ‘video’”) + pricing highlight | conversion > likes; CTAs crank follower→client funnel |
| hook strength | slow fade‑in → 0‑2 s punch, clear concept | completion rate & replays dominate reel ranking |
| trend leverage | obscure track → trending audio + remix sticker | surfing trend boosts discoverability; recency is a top signal |
| social proof density | no comments → artist shout‑outs, collab tags, fan duets | external validation both ranks & persuades |
| format hygiene | 16:9, long captions → 9:16, 6‑15 s, native edits, closed‑caps | algorithm favors “made‑for‑reel” metadata |
| network activation | passive post → “tag the band you’d cast here” prompt | comments + shares multiply reach weight |
| portfolio breadth | single style → carousel of radically different looks | shows adaptability, dissolves “can they do MY vibe?” objection |
You can then ask for candidate videos (or provide your own) and have the LLM rank by dimension:
| idea | authority | genre_fit | commissionability | hook_strength | trend_leverage | social_proof | format_hygiene | network_activation | portfolio_breadth |
| 30s reel: behind‑the‑scenes glam rock shoot w/ artist tag + shotlist overlay | 4 | 4 | 3 | 3 | 2 | 4 | 4 | 3 | 2 |
| 12s reel: fast‑cut genre demo montage (pop, indie, hiphop) w/ trending sped‑up audio | 2 | 5 | 3 | 5 | 5 | 3 | 4 | 4 | 5 |
| 15s testimonial: happy client artist talking on‑cam, intercut finished shots, clear ‘dm to book′ | 5 | 4 | 5 | 4 | 2 | 5 | 4 | 3 | 2 |
| 8s hook: extreme macro sparkle shot ending w/ logo reveal, trending glam track | 3 | 4 | 2 | 5 | 4 | 2 | 4 | 2 | 1 |
| carousel: storyboard frames + final shots for recent commission, caption explains process | 4 | 3 | 4 | 3 | 1 | 3 | 3 | 2 | 3 |
| reel: split‑screen ‘before vs after’ color grade on mobile footage vs final cut | 3 | 3 | 4 | 4 | 2 | 3 | 4 | 2 | 3 |
| story highlight: pricing tiers infographic + link sticker to inquiry form | 4 | 3 | 5 | 2 | 1 | 3 | 3 | 2 | 1 |
| 15s concept pitch: ‘which band fits this visual?’ prompt, encourages tagging | 2 | 4 | 4 | 4 | 3 | 4 | 3 | 5 | 3 |
| 9s reel: top 3 camera angle tricks demoed w/ on‑screen text | 3 | 3 | 3 | 4 | 3 | 2 | 4 | 2 | 4 |
| 12s reel: collab remix with popular cinematographer duet, shows adaptability | 4 | 4 | 3 | 4 | 4 | 5 | 4 | 4 | 4 |
If you don’t like the dimensions or the examples, you can ask for new ones, or pick a different choice to dimensionalize.
Hope this is helpful!
The central idea (avoid single-factor models, while acknowledging this is itself just a single factor among many) is entirely mine.
Several of the sub-threads, such as Structural Defenses (e.g. non-AI methods to prevent one-shottedness), the Activist-Ideologue distinction, and Metastasization of the Local Frame were suggested during discussion with Claude and GPT5 about the central idea, and I chose to incorporate them after further discussion. Many potential sub-threads did not make the cut.
Others of the sub-threads, such as Why This Persists, the Self-Own, and and the general importance of factors and difficulty of multi-factor thinking are my ideas that were deepened through discussion with AI.
Dimensionalization and Conceptual Rhyme are my ideas that I have written previously about how to get AI to perform on command; I developed these working closely with AI.
Some phrasing is AI-written, but most of the writing is my own.
This is the first essay in what I intend to be a brief series about using AI to help automate and scale perspective-taking.