SYNTHESIS NOTE
TopicsReasoning Logic Internal Rulesthis note

Can LLMs reason creatively beyond conventional problem-solving?

Explores whether large language models can engage in truly creative reasoning that expands or redefines solution spaces, rather than just decomposing known problems. This matters because existing reasoning methods may miss creative capabilities entirely.

Synthesis note · 2026-02-22 · sourced from Reasoning Logic Internal Rules
What makes chain-of-thought reasoning actually work? How do you navigate synthesis across fragmented research topics? What kind of thing is an LLM really?

The Universe of Thoughts (UoT) paper identifies a blind spot in the LLM reasoning literature: all existing methods (CoT, ToT, GoT, Forest-of-Thought) focus on conventional problem-solving — decomposing known problem types into manageable steps. None address creative reasoning, where the solution space itself must be expanded or redefined.

Drawing on Boden's established cognitive science framework, UoT defines three creative reasoning paradigms:

1. Combinational Creative Reasoning: Identifying solutions from other domains that are relevant to the target problem but have not been previously applied there. The mechanism: cross-pollination of known solutions across domain boundaries. A collage is combinational — existing visuals arranged in unconventional ways.

2. Exploratory Creative Reasoning: Adopting individual building blocks (not solutions) from outside the target solution space. New conceptual primitives expand what's possible within the existing framework. Impressionism was exploratory — brushstrokes used in a functionally new way within painting's existing rules.

3. Transformational Creative Reasoning: Fundamentally altering or dropping the core rules that define the solution space. This changes what solutions are even conceivable. Cubism was transformational — breaking the rule of direct representation to depict objects from multiple angles.

The hierarchy is important: combinational reuses, exploratory expands, transformational redefines. Each requires progressively deeper deviation from conventional reasoning patterns.

UoT introduces evaluation metrics orthogonal to standard reasoning benchmarks: feasibility as a constraint (creative solutions must still be implementable), with utility and novelty as metrics. This three-axis evaluation addresses a gap identified by Can LLMs generate more novel ideas than human experts? — LLMs can generate novel outputs but cannot evaluate their own creativity.

The connection to Why do LLMs generate novel ideas from narrow ranges? is direct: diversity collapse may occur precisely because existing reasoning methods explore only one paradigm (combinational at best) while neglecting exploratory and transformational modes. Explicitly prompting for each paradigm could address the diversity problem.

Inquiring lines that read this note 40

This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.

Why do LLM research ideas score high on novelty yet collapse into low diversity? How do language models inherit human biases from training data? Can prompting strategies overcome LLM biases without model fine-tuning? When does optimizing for quality undermine the value of diversity? Why can LLMs generate ideas better than they evaluate them? How does reasoning graph topology affect breakthrough insights and generalization? Can prompting inject entirely new knowledge into language models? How faithfully do LLMs reflect their actual reasoning in outputs and explanations? Do language models understand semantics or rely on pattern matching? How can AI agents autonomously learn and transfer skills across tasks? Do language models perform faithful symbolic reasoning independent of semantic grounding? How does policy entropy collapse constrain reasoning-focused reinforcement learning? Why do reasoning models fail at systematic problem-solving and search? What capability tradeoffs emerge when scaling model reasoning abilities? How do we evaluate AI systems when user perception misleads actual performance? Does reinforcement learning teach reasoning or just when to reason? How effectively do deterministic tools improve language model reasoning on formal tasks?

Related concepts in this collection 4

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
15 direct connections · 136 in 2-hop network ·dense cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

creative reasoning requires three distinct paradigms — combinational exploratory and transformative — that existing reasoning methods do not address