C-UoT (combinational):

STEP 1: Find analogous problems from other domains → list of problems
STEP 2: Decompose the analogs' solutions into components ("thoughts") → component pool
STEP 3: Pick a "host solution" from the target domain → base solution
STEP 4: Find components to swap (functionally similar, outwardly different) → swap candidates
STEP 5: Plug in components from the analogs → new solution
STEP 6: Score on usefulness, novelty, feasibility → ranked list

T-UoT (transformational):

STEP 1: Identify the problem's explicit rules → list of rules
STEP 2: Uncover hidden assumptions → expanded list of rules  
STEP 3: Find analogous rules from other domains → rule pool
STEP 4: Mutate the rules (remove / change / add) → new rule sets
STEP 5: Apply E-UoT in each new space → creative solutions

Every step is a sequential prompt to the LLM. You can run it all in one conversation, giving instructions step by step.

Task: A coffee shop in a residential suburb of Chicago. Foot traffic is close to zero and rent eats the margin. Standard fixes (ads, discounts, delivery) don't work because competitors are doing the same thing. You need a creative solution.

Prompt (T-UoT — transformational):

I want to find a creative solution to a business problem. Use the transformational creative thinking method.

PROBLEM: Increase revenue for a coffee shop in a residential suburb with low foot traffic.

CONSTRAINTS:
- Minimal marketing budget
- No option to change location
- Can't operate at a loss

STEP 1: Identify ALL the rules I treat as fixed (explicit rules and hidden assumptions).

STEP 2: For each rule, find an analogous rule from a completely different domain (not food service) that serves the same function but looks different.

STEP 3: Propose 3 rule mutations:
- REMOVE one rule (which one and why)
- CHANGE one rule to an analog from another domain
- ADD a rule from another domain

STEP 4: For each modified rule set, generate a specific solution.

STEP 5: Score each solution on a 1-10 scale: usefulness, novelty, feasibility.

Result: The model will first surface hidden assumptions: "customers have to come in on their own", "a coffee shop is a place to drink coffee", "revenue = selling drinks". Then it will find analogs from other domains: the subscription model (SaaS), mobile points of sale (food trucks), coworking spaces, club formats. The rule mutations will yield solutions like: "the coffee shop as an Amazon package pickup point + free coffee", "a morning coffee subscription delivered to apartment building lobbies", "renting out the space for small neighborhood events". At the end, a ranked list with scores.

The LLM's weakness: Models are trained on patterns, so they generate typical solutions. Ask one to solve a business problem and you'll get a mashup of Medium posts and Hacker News threads. That's not creativity, that's averaging.

The LLM's strength: Models hold a huge amount of knowledge from many domains and can spot structural analogies. They can see that the problem of "a single-lane bridge with traffic jams" is structurally similar to "an elevator in a skyscraper" or "an airport runway". That's the foundation for transferring ideas across domains.

How UoT uses this: Instead of "come up with a solution", the prompt steers the model down a specific path:

1. First find analogous problems (structurally similar, but from other fields)
2. Break down their solutions into components
3. Plug those components into your problem
4. For maximum creativity, change the rules of the problem itself

This turns the search for a solution from "guess a good answer" into a systematic exploration of the space of possibilities.

Control knobs:

• Number of analogous domains: more = more diverse ideas, but more "noise"
• Depth of decomposition: smaller components = more combinations
• How radical the rule mutations are: removing a rule vs. replacing it with an analog gives a different level of "craziness" in the solutions

C-UoT (combinational) — when you need fresh ideas without radical changes:

Use combinational creative thinking to solve the problem.

PROBLEM: {problem_description}
CONSTRAINTS: {list_of_constraints}
CURRENT SOLUTION (if any): {base_solution}

STEP 1 — ANALOGIES:
Find 3-5 problems from OTHER domains (not {current_domain}) that are structurally similar to my problem but look completely different.

STEP 2 — DECOMPOSITION:
For each analogous problem, describe 1-2 known solutions and break them down into key components (mechanisms, principles, elements).

STEP 3 — TRANSFER:
Pick the 3-5 most interesting components from other domains. Explain what function they serve and how they could be adapted to my problem.

STEP 4 — SYNTHESIS:
Create 3 new solutions by combining components from different domains with elements of my problem.

STEP 5 — EVALUATION:
Score each solution (1-10): usefulness, novelty, feasibility. Explain the scores.

T-UoT (transformational) — when you need a breakthrough, "thinking outside the box":

Use transformational creative thinking to solve the problem.

PROBLEM: {problem_description}
CONSTRAINTS: {list_of_constraints}

STEP 1 — RULES (explicit):
List all the explicit rules and constraints of the problem.

STEP 2 — RULES (hidden):
Identify the hidden assumptions I take for granted that could actually be changed. What do I consider "the way it's done", "the way it's always been", "obvious"?

STEP 3 — ANALOGOUS RULES:
For each rule (explicit and hidden), find an analogous rule from another domain that serves the same function but works differently.

STEP 4 — MUTATIONS:
Create 3 new rule sets:
A) REMOVE one key rule — what becomes possible?
B) REPLACE one rule with an analog from another domain
C) ADD a rule from another domain that wasn't there before

STEP 5 — SOLUTIONS:
For each new rule set, generate a specific solution that would have been impossible under the old rules.

STEP 6 — EVALUATION:
Score the solutions (1-10): usefulness, novelty, feasibility. Which solution do you recommend and why?

Placeholders:

• {problem_description} — the specific problem you're solving
• {list_of_constraints} — what can't change (budget, deadlines, resources)
• {current_domain} — your problem's field (so the analogies come from other fields)
• {base_solution} — your current approach, if any (for C-UoT)

🚀 Quick start — paste into the chat:

Here is a transformational creative thinking template. Adapt it to my problem: [your problem]. Ask me questions to fill in the fields.

[paste the T-UoT template above]

The LLM will ask about your constraints and current domain, which it needs in order to look for analogies outside your field and understand which rules can be mutated.

⚠️ Novelty vs. usefulness: E-UoT (exploratory) produces the most novel ideas, but they're often less practical. In the experiment, E-UoT showed novelty of 0.81 but usefulness of only 0.28 on one of the tasks. The more radical the creativity, the harder it is to bring down to earth.

⚠️ Not for simple problems: UoT is overkill for problems with a clear answer. Use it for ill-defined problems: strategy, innovation, product design, creative projects.

⚠️ Requires iteration: One prompt won't work magic. The method assumes sequential steps with intermediate results. Be ready for 5-10 back-and-forths with the model.

⚠️ Analogy quality depends on the model: Weak models find shallow analogies. For the best results, use GPT-4, Claude, or Gemini Pro.

The researchers built three tasks where standard solutions are known to fall short: optimizing a single-lane bridge (you can't build a new one), electricity pricing (you can't add capacity), and strengthening social ties in a community (innovative interventions required). Each task has a huge solution space and demands creativity.

The key methodological innovation is canonicalizing solutions before scoring. Different methods produce answers at different levels of detail: one gives a one-sentence idea, another a full plan. To compare apples to apples, every solution was first compressed to its core idea with a separate LLM call.

Metrics: usefulness (how well it solves the problem, from 0 to 1), novelty (cosine distance from known solutions in embedding space), feasibility (whether constraints are respected). Creativity = usefulness + novelty for solutions that clear the feasibility threshold.

The results were surprising: T-UoT on GPT-4o (0.698) beat GPT-5 (0.649) on the bridge task. This shows that reasoning structure matters more than raw model power. The method squeezes more creativity out of a "weaker" model than a "stronger" model manages without it.

An interesting finding: as the method gets more complex (C→E→T), novelty goes up but usefulness can drop. T-UoT generates the most radical ideas but, at the final generation step, often picks a more conservative solution: a balance between "craziness" and practicality.

Context: An example of C-UoT on the single-lane bridge optimization task. The researchers show how the method finds analogies and transfers components.

We first consider analogous single-channel problems—single-track railway dispatch, 
elevator up-peak control, runway slot assignment, and appointment booking systems. 
We decompose their solutions into portable thoughts: 
(i) time-slot reservations / tokens to regulate access, 
(ii) platooning/batching to cut direction switch losses, 
(iii) priority windows for critical classes, 
(iv) ramp-metering at entries, and 
(v) time-shift incentives to push discretionary trips off peak. 

Starting from a naïve host policy ("switch direction whenever a car arrives"), 
we target two high-impact slots—the direction schedule and the arrival interface—
and swap in far-but-apt components: a tokenized pre-booking system with 5-minute slots 
(from runways/clinics), batch service before each switch (from elevators), 
and off-peak credits for grocery trips (from appointment economics).

T-UoT example — surfacing hidden rules:

T-UoT starts by surfacing hidden rules—e.g., people must cross in person, 
groceries require a shopper trip, only road vehicles are allowed, 
and access is first-come-first-served (FCFS). 

The transformational step is to transform these rules: 
- permit supervised non-road corridors (water/air), 
- allow goods to move without people (drone lockers, mobile markets), 
- or replace FCFS with credit allocation. 

Each transformed rule set allows for solutions that were impossible before: 
drone delivery eliminates many grocery trips; supervised air/water shuttles 
add a parallel channel; quota credits reserve scarce peak crossings for essential work.

💡 Adaptation for personal decisions:

UoT works great for career and life crossroads where standard advice doesn't help.

Use transformational creative thinking.

PROBLEM: I'm 35, a middle manager at a bank. I'm burned out and want to change careers, but I don't know to what. I have a mortgage and two kids.

STEP 1: What rules do I treat as fixed in my career situation?

STEP 2: What hidden assumptions am I making? ("I need a stable salary", "Changing careers = starting from zero", "At 35 it's too late to learn")

STEP 3: Find examples of people/situations from other fields where similar "rules" were broken with good results.

STEP 4: Propose 3 options where I change one of my rules. What becomes possible?

🔧 Technique: add a "devil's advocate" at the evaluation stage

In the original method, evaluation uses three metrics (usefulness, novelty, feasibility). You can tighten the filtering:

STEP 6 — CRITIQUE:
For each solution, answer:
- Why WON'T this work? (3 reasons)
- What needs to change to eliminate these risks?
- After revising, update the scores.

This forces the model to bring radical ideas down to earth until they're feasible.

Paper: "Universe of Thoughts: Enabling Creative Reasoning with Large Language Models"
Authors: Yuto Suzuki, Farnoush Banaei-Kashani (University of Colorado Denver)
Theoretical foundation: Margaret Boden — "The Creative Mind: Myths and Mechanisms" (2004)
Status: Preprint