Conceptual Rhyme and Metaphor

Systematic thinking transfers

tl;dr

Most new problems are not completely novel. A fast way to get traction in a new domain is to transfer working know-how from an adjacent domain, while being careful not to smuggle in bad assumptions.

That’s what this stack is for.

• Rhyme: identify two ideas that share patterns

  • ex. “a game of football rhymes with trench warfare (infantry battle for territorial gain under attrition)”

• Metaphor: build an explicit map from one domain to another

  • ex. “coach : HQ, quarterback : platoon leader, front line : line of scrimmage, end zone : combat objective, …”

These mental moves are cognitively taxing, for me at least, so historically I have limited my use of them to complex situations that reward both rigor and creativity. Now, LLMs can automate them well enough to make the moves cheap, fast, and repeatable.

Programming note

Always use the Thinking/Reasoning mode in the LLM of your choice when running any prompts on this blog.

Definitions, knobs, and templates now live in the brand-new FUTURE TOKENS GitHub. This post focuses on how to use the Rhyme/Metaphor stack in practice, with concrete examples.

---

Rhyme

A few things to note:

• We are rhyming concepts, not sounds. Just as lyrical rhymes have shared characters and sounds but are not identical words, so rhyming concepts have shared parts despite some differences.

• An ideal concept to try to Rhyme is one where we are looking to learn, teach, change our approach towards, or try new things within.

• We are not looking for perfection or Truth, just enough parallelism to be worth exploring further.

Rhyme starts with one concept and generates a shortlist of concepts that have interesting parallels.

Suppose I want to understand why people spend a lot of time on TikTok.

Try this prompt:

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/rhyme.md

Then rhyme "scrolling TikTok".

1. Panning for gold in a fast stream: high-throughput flow, sparse jackpots, hand motions that sift, skill mostly in “where/when to swirl,” not in producing gold.

2. Patch foraging in ecology: you probe “content patches” and decide when to leave; rewards deplete locally; travel cost is time/attention.

3. Industrial trawling with bycatch controls: net a ton, keep a fraction; quality hinges on mesh and depth.

4. Scanning a crowded radio band with squelch: signal-to-noise, channel hopping, squelch thresholds, intermittent clear signals.

5. QC on a fast assembly line: rapid yes/no triage, rare exceptions get pulled for deeper inspection, fatigue increases misses.

Full output below.¹

I am partial to “river panning for gold”. We don’t have enough positive framings for TikTok, considering how much people seem to like it.

Other examples below.²

Since we seem to be on an interesting track, we can go deeper.

---

---

Metaphor

Assume we have a Rhyme we like, such as “scrolling TikTok”↔“panning for gold in a river”. What would it mean if we imported all of the structure of the concept “panning for gold in a river”, and mapped everything we reasonably can to the concept “scrolling TikTok”?

Metaphor starts with two concepts and generates a complete list of parallels between them.

Try this prompt:

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/metaphorize.md

Then metaphorize "scrolling TikTok is panning for gold in a river".

Now we have a fairly dense framework for thinking about TikTok through the lens of gold panning.

It’s a fine map, not amazing. Gold panning is not a rich domain; there is not much theory behind it.

Metaphorization becomes extremely powerful when math can port from one domain to another, which we don’t have in this case. To see that, try “social media is applied virology” or “project management is container port operations”.³

Metaphorization works with any metaphor (or simile, analogy, etc.).

• Failure is the condiment that gives success its flavor. (Truman Capote)

• A computer is a bicycle for the mind. (Steve Jobs)

• All that is transitory is but a metaphor. (Goethe)

In any case, Metaphorization outputs a map. There are several high-leverage uses of such a map that I have found:

• Importing functions

• Searching for missing parallels

• Catching failure modes

Importing functions

If the map is good, then known-good ways of thinking about one domain should work in the other. In other words, the map might help prevent us from overfitting to domain-specific phenomena.

What approaches work in both "panning for gold" and "scrolling TikTok"?

Some highlights:

1. Define the spec up front

   S: “Pay dirt” = ≥X g/hr.

   T: “Gold” = clips that you Save/Note because they teach/seed ideas/relax. Write the rule.

2. Move-on rule (marginal value theorem)

   S: Leave when instant yield < travel-adjusted average.

   T: If last ~8–10 swipes produce 0 saves, hop topic/creator; don’t chase sunk time.

3. Tighten the mesh

   S: Use finer classification to dump silt faster.

   T: Aggressive “Not Interested”/mute/block; shorter trial window (e.g., 2–3s) before skipping.

We might improve our experience on TikTok by using the same strategies that work in gold panning.

Per above, we can attempt to define a threshold that indicates the feed is good, move on when it’s not, and use the tools available to us to filter for what we want.

Searching for missing parallels

If the map is good, then existing structure in one domain might inspire the creation of missing structure in the other. In other words, the map itself might generalize.

What structures are present in "panning for gold", but absent from "scrolling TikTok"?

Some highlights:

1. Objective payoff. Gold has a market price; TikTok’s “value” is subjective (amusement, ideas, leads).

2. External constraints. Daylight, weather, permits, fuel impose hard stops; TikTok is 24/7 with no natural braking.

3. Locality & travel cost. Moving bars/claims is costly and path-dependent; switching topics/creators is nearly free.

We might improve our experience on TikTok by taking on some of the baggage of gold prospecting, even where it doesn’t seem to strictly apply.

Per above, we can attempt to make the payoff of scrolling objective (e.g. always share finds with friends), constrain our usage hours, or quit and wait before we switch feeds.

Catching failure modes

If the map is good, then what is known to fail in one domain might fail similarly in the other. In other words, the map can tell us where not to go.

What approaches fail in "panning for gold" that are likely to also fail in "scrolling TikTok"?

Some highlights:

1. Sunk-cost squatting on a barren bar.

   Staying because you already dug is dead time; doom-staying on a weak topic/creator is the same.

2. Chasing glitter (fool’s gold).

   Shiny mica ≠ gold; glossy clickbait ≠ value.

3. Prospecting forever, never sluicing.

   All sampling, no production; all browsing, no drilling profiles/hashtags.

We might improve our experience on TikTok by avoiding the same strategies that don’t work in gold prospecting.

Per above, we can move on when we’re no longer enjoying a topic, make sure our enjoyment is genuine, and follow specific creators rather than remain in the torrent.

Strategy

The above represents the beginnings of a strategy for better enjoying TikTok scrolling, inspired by the experience of gold prospectors. The framing is far from perfect, yet it can still be useful.

Using the prompts from FUTURE TOKENS, we can create a similar map for virtually any pair of concepts.

---

Coda

Rhyme and Metaphor have always been powerful seeds for ideation. Now we can use LLMs to automate and scale this process.

I’d love to hear about it if you find interesting ones!

1

1. Panning for gold in a fast stream

• Why it rhymes: high-throughput flow, sparse jackpots, hand motions that sift, skill mostly in “where/when to swirl,” not in producing gold.

• Ports: set a “mesh size” (Not Interested aggressively), swirl (re-watch once to test value), relocate when pan yields 5–7 duds in a row, keep a vial (Save list) for later refining.

• Doesn’t port: rivers don’t adapt to you; TikTok’s stream does—so your actions change the seam you’re mining.

2. Patch foraging in ecology (marginal value theorem)

• Why it rhymes: you probe “content patches” and decide when to leave; rewards deplete locally; travel cost is time/attention.

• Ports: leave a patch when instant reward rate falls below your session average minus switch cost; pre-choose target prey (tags/creators) to cut bycatch; use “Follow” as a beacon to re-find rich patches.

• Doesn’t port: content patches can regenerate instantly via the model; real patches don’t.

3. Industrial trawling with bycatch control

• Why it rhymes: net a ton, keep a fraction; quality hinges on mesh and depth.

• Ports: tune mesh (mute/block, long-press Not Interested), fish at the right depth (niche hashtags), respect quotas (timebox); land high-value fish into chilled hold (Collections).

• Doesn’t port: trawling damages the seabed; scrolling doesn’t—your harm is cognitive fatigue and model drift.

4. Scanning a crowded radio band with squelch

• Why it rhymes: signal-to-noise, channel hopping, squelch thresholds, intermittent clear signals.

• Ports: raise squelch (minimum watch time before like/save), scan with cadence (3–5 seconds then flip), lock when carrier is strong (creator fits a theme), log frequencies that pay (search terms).

• Doesn’t port: radio spectra are stationary; the FYP re-tunes to your scans.

5. QC on a fast assembly line

• Why it rhymes: rapid yes/no triage, rare exceptions get pulled for deeper inspection, fatigue increases misses.

• Ports: define defect criteria (what you refuse to watch), add stop-the-line moments (hard pause to decide: save/share/comment or dump), rotate inspectors (switch themes) to fight drift.

• Doesn’t port: QC optimizes against specs; TikTok’s “spec” is your evolving taste—move the spec deliberately.

6. Tape-reading in a hot market

• Why it rhymes: continuous tape, micro-setups (hooks, cuts), variable-ratio reinforcement, risk is overtrading.

• Ports: pre-name setups you want (educational hooks, demo-then-explain), use hard stops (session cap), scale into themes that keep printing (make playlists; follow adjacent creators), journal winners (why a clip worked).

• Doesn’t port: markets push back via liquidity/impact; TikTok pushes back via recommendation shift and novelty decay.

2. Other examples of Rhymes

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/rhyme.md  

Then rhyme "playing strategy games".

• Running a pop-up kitchen on a slammed night

  • Prep & mise en place = early eco/build order.

  • Ticket queue flow = lane pressure; expo sets tempo.

  • Limited specials = denial plays vs competitors’ staples.

• Solo morning with two small kids

  • Bottlenecks (shoes, teeth) drive sequencing like build orders.

  • Mood scouting removes fog-of-war; snacks = power spikes.

  • Parallelize safely; reserve attention as emergency mana.

• DJing a dance floor

  • Warm-up tracks = economy; bangers = timing pushes.

  • Reading the room = reconnaissance; teasers = feints.

  • Energy waves control tempo; avoid overextending the crowd.

• Permaculture gardening

  • Soil health snowballs like compounding advantages.

  • Water/light are resource nodes; pests are raids.

  • Succession planting = tech tree; guilds = unit synergies.

• Airport ramp ops in a thunderstorm

  • Gates/equipment are hard caps; de-icing windows = attack timers.

  • Reroutes = flanks; ground stops = crowd control.

  • Info delays are fog; clear comms = map hacks.

• ER triage during a surge

  • Beds/staff scarce → targeting priorities.

  • Diagnostics = scouting; protocols = rehearsed build orders.

  • Escalate/de-escalate as counter-tech adoption.

• Multiparty dealmaking

  • Coalition math = alliance meta; walk-away = denial threat.

  • Concessions are resource trades; expiries create timing windows.

  • Signaling and feints shape the negotiation map.

• Live incident response (infosec)

  • Patch order = build order; budget/time vs threat surface.

  • Telemetry = scouting; honeypots = traps/feints.

  • Attacker’s privilege ladder vs your detection tech tree.

• Thru-hike route planning

  • Rations = economy; campsites = expansions.

  • Weather windows = timing attacks; bailout routes = retreats.

  • Pack weight = supply cap; tempo hinges on encumbrance.

• Editing a feature film

  • Footage inventory = resource pool; sequences = lanes.

  • Beats = timing pushes; foreshadowing denies later confusion.

  • Late recuts = tech debt; lock when marginal value drops.

• Startup go-to-market

  • Beachhead segment = first expansion.

  • Burn vs growth = eco vs army.

  • Pricing tiers = unit tiers; partnerships = alliances.

• Orchestra rehearsal

  • Section drills (micro) vs balance/form (macro).

  • Rehearsal minutes allocate to bottlenecks (highest leverage).

  • Tempo/dynamics = map control; solos = power spikes.

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/rhyme.md

Then rhyme "having a second kid".

1. Single-threaded → multi-threaded household

   Shared memory (kitchen, calendar) creates contention; you need locks (rituals), semaphores (signals), and queue discipline (handoffs). Useful for designing “no-conflict” zones/times.

2. Dual-core CPU + scheduler

   Load balancing between parents; preemption (newborn interrupts); watch for priority inversion (urgent toddler need starved by constant infant I/O). KPIs: backlog size, swap time between tasks.

3. Two oscillators → beat interference

   Sleep/wake cycles with slightly different periods create constructive/destructive interference. Hunt for resonance (aligned naps) and dampeners (white noise, buffer windows).

4. Zero-downtime deploy behind a load balancer

   You can’t take Kid #1 offline while #2 “deploys.” Health checks, rolling updates to routines, graceful degradation for noncritical chores.

5. Sharded datastore with conflict-free merges (CRDT vibes)

   State lives across caregivers/rooms; eventual consistency beats perfect sync. Define merge rules (if two versions of “who promised what,” prefer written plan over memory).

6. From two-body to n-body dynamics

   Nonlinear coupling (parent ↔ kid ↔ sibling). Fewer closed-form “laws,” more heuristics and simulation. Good warning against overfitting Kid #2 to Kid #1 playbook.

7. Small stovetop, second hot pan

   Limited burner real estate (time/attention). Mise en place (prep at night), batch cooking (errands in clusters), and pan-swapping (tag-team coverage) matter more.

8. A startup adding a second product line while the first scales

   Shared brand and ops, different user needs. Avoid regression (don’t break Kid #1’s stable routines); reuse platforms (bedtime scripts), fork UX where necessary.

9. New hire joins a tight two-person team

   Storming→norming. Roles renegotiated, bus factor rises (fewer single points of failure). Plan explicit “one-on-one”s (parent↔Rune; parent↔Rava) to stabilize culture.

10. Two train lines sharing a junction

    Crunch windows (mornings, bedtimes) are the interlock. Add sidings (staging areas), fixed dwell times (non-negotiable buffers), and dispatch rules (who yields when).

11. Portfolio of two correlated assets

    Shocks spill over (sleepless night ↔ workday). Hedge with reserves (sleep/meal buffers), rebalance attention, and accept purposeful tracking error (fair ≠ equal).

12. ABAB vs AABB coverage patterns

    Formal scheduling motif: alternate nights (ABAB) for recovery, or cluster (AABB) for depth. Switch when interference patterns shift.

Quick do-not-ports (exclusions): kids aren’t fungible tasks; “throughput” metaphors can hide feelings; fairness ≠ identical treatment; sibling identity isn’t a copy.

3. Other examples of metaphors

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/metaphorize.md

Then metaphorize "project management is container port operations"

1) Entity Mapping (Port → Project)

• Vessels arriving → Workstreams/releases arriving

  Each has ETA, priority, and cut-off windows.

• Berths → Concurrent work slots / WIP limit

  Scarce capacity; over-berthing = overcommit.

• Quay cranes → Specialized people/teams/tools

  Setup/changeover costs matter.

• Yard blocks → Queues/backlogs by domain

  Every rehandle = thrash/rework.

• Customs holds → Blocking dependencies/approvals

  Invisible until declared; track explicitly.

• Weather/strikes → Exogenous risk

  Absorb with buffers; don’t “schedule through.”

• Truck gates/appointments → Intake gating / start controls

  Control arrival rate to stabilize flow.

• Vessel turn time → Milestone cycle time

  Arrival → ready → worked → departed (accepted).

2) State Variables (track both sides)

• Throughput (λ): containers/day ↔ tasks accepted/day.

• Capacity (μ): crane moves/day ↔ effective team service rate/day.

• Utilization (ρ): ρ = λ/μ at the bottleneck. Target ρ ≤ 0.8.

• WIP (L): boxes in yard/berth ↔ tasks in progress.

• Cycle time (W): arrival→departure ↔ start→done.

• Rehandle factor (R): extra lifts/box ↔ extra touches/task.

• Hot ratio (H): % expedites (“hot” containers) ↔ % urgent work.

• Dwell mix: wait + work + hold ↔ queue + active + blocked.

3) Control Policies to Port

1. Berth = WIP discipline

   • Set hard WIP limit = number of berths. New work waits offshore (backlog) until a berth frees.

   • Keep one expedite berth for true preemption.

2. Arrival gating (truck gate → intake gate)

   • Appointment windows for starts; limit daily starts to hold ρ stable.

   • Stagger big inflows (no “five vessels at once”).

3. Crane assignment (teams to value)

   • Put the best crane on the bottleneck vessel (critical path).

   • Minimize changeovers: cluster similar work.

4. Yard plan (queue design)

   • Separate queues by class (core vs spikes vs research).

   • Place high-rehandle items near integration (“the pass”).

5. Customs first (unblock early)

   • Identify likely holds (legal, security, data access) pre-arrival; clear before berthing.

6. Weather buffers

   • Maintain 20–30% slack at the bottleneck.

7. Cut-off & sail (integration cadence)

   • Fixed cut-offs: during merge windows accept no late cargo. Ship what’s on the quay; defer the rest.

Read https://raw.githubusercontent.com/jordanrubin/FUTURE_TOKENS/refs/heads/main/metaphorize.md

Then metaphorize "social media is applied virology"

Source (virology) to target (social media) mapping

A virion in virology corresponds to a post in social media: a discrete, packaged unit meant for transmission. The genome of the virion is like the core message or meme inside the post — the payload that determines replication instructions. A capsid, the protein shell that protects the genome, becomes the format and framing of the post — the headline, image, or style that keeps the message intact until delivery.

Envelope proteins, which bind to host cells, are equivalent to the affordances of a platform — things like hashtags, mentions, thumbnails, and push notifications — that determine whether the post successfully attaches to a user’s attention. The host cell is the user’s mind, with varying susceptibility depending on “immune” factors like skepticism or boredom.

An infection in virology maps to engagement — when the user actually consumes and registers the content. The viral hijacking of the host’s replication machinery corresponds to the user resharing, commenting, or otherwise propagating the content, driven by emotional triggers that act like hijacking proteins.

The viral load is the feed saturation of a given meme, with high levels increasing spread probability but also risking burnout. The mutation rate in viruses becomes the rate of meme adaptation — new captions, remixes, or formats that help evade “immune” fatigue. The incubation period is the delay between initial shares and mass visibility, often shortened by algorithmic amplification. Transmission vectors — like air or blood — become different sharing channels such as direct messages, public feeds, and group chats.

The basic reproduction number (R₀) is the viral coefficient (K) — the average number of new sharers each existing sharer produces. The immune system is skepticism and moderation algorithms. Herd immunity is meme fatigue, and pathogenicity becomes the emotional or behavioral impact of the post — with high severity potentially leading to faster burnout. Finally, epidemiological models map directly onto engagement growth models.

---

Ported rules and heuristics

Transmission thrives in high-contact environments — meaning that algorithms maximizing exposure increase K. Mutation can evade immunity, so variants of a meme can reignite engagement after fatigue. Over-virulence can burn out the host, just as overly manipulative or polarizing content may cause users to disengage or leave a platform. Quarantine slows spread — in social media this means content suppression, deboosting, or locking it in closed groups. Vector-specific adaptation matters — a meme optimized for Twitter may fail on TikTok. Superspreaders dominate distribution — influencers can have a disproportionate effect. Silent carriers — lurkers who quietly forward content — can drive stealth outbreaks.

---

Quantitative port using the SIR model

In epidemiology, the SIR model tracks the susceptible (S), infected (I), and recovered (R) populations. In social media, these are the unexposed users (U), the engaged sharers (E), and the fatigued or immune users (F). The transmission rate β is the probability that a contact between a sharer and an unexposed user leads to engagement. The recovery rate γ is the rate at which sharers stop resharing. The total population N is the number of reachable users. The viral coefficient K is β divided by γ — if K is greater than 1, the content grows exponentially until constraints hit.

The core SIR equations become:

dU/dt = -β U E / N

dE/dt = β U E / N - γ E

dF/dt = γ E