Examining and Addressing Barriers to Diversity in LLM-Generated Ideas

Summary

Identifies two distinct mechanisms suppressing diversity in LLM ideation: individual-level fixation, where early outputs constrain subsequent ideation, and collective-level failure of knowledge partitioning, because LLMs aggregate knowledge into one unified distribution rather than occupying distinct regions of the knowledge space as human populations do. Four studies show CoT prompting fixes fixation and ordinary (rather than celebrity-creative) personas fix partitioning, and combining both produces the highest diversity, outperforming humans. Supplies both a theoretical vocabulary and a measurement approach for the paper's exploration/novelty outcome, from a management and marketing framing rather than a pure NLP one.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a high-priority candidate for the Architecture as Epistemology (ISR) paper (angle: epistemic quality of LLM outputs; correlated errors / homogeneity across LLM agents). Verified against: Opened https://arxiv.org/abs/2602.20408 (arXiv abstract page, v1 23 Feb 2026)

Critique / open questions

Queued — not yet assessed.

Key quotes

Ideas generated by independent samples of humans tend to be more diverse than ideas generated from independent LLM samples, raising concerns that widespread reliance on LLMs could homogenize ideation and undermine innovation at a societal level. (abstract)

Drawing on cognitive psychology, we identify (both theoretically and empirically) two mechanisms undermining LLM idea diversity. (abstract)