Anchorless Diversification for Parallel LLM Ideation

2026-05-28

AI Summary

This work addresses the challenge of enhancing semantic diversity in parallel creative generation with large language models while maintaining output quality and computational efficiency. The authors propose a seed-free, inference-time control method that integrates independent generation, hierarchical semantic direction modeling, and group-referenced divergence instructions, enabling coordinated multi-directional high-quality generation with only a single planning call. Experimental results demonstrate that group-referenced divergence serves as a highly effective low-cost baseline that substantially boosts diversity. Furthermore, the proposed semantic direction hierarchy achieves a superior trade-off among diversity, generation quality, and computational overhead compared to conventional anchor-based regeneration approaches, whose advantages notably diminish when accounting for end-to-end token costs.

Abstract

LLMs are increasingly used to generate candidate-idea pools for creative tasks where broad exploration is valuable. Parallel inference can be attractive in this setting when it broadens the pool while retaining quality and cost efficiency. We study inference-time controls for candidate-pool diversification, asking whether anchorless methods can rival methods that depend on observed seed ideas. Across three creative task families, we compare independent generation and semantic direction stratification with self-, peer-, and representative-anchor baselines, under neutral and population-referential divergent instructions. Population-referential divergence is a strong low-cost baseline, increasing semantic diversity while preserving quality proxies. Semantic direction stratification is stronger: a single planning call organizes generations across broad semantic directions, yielding the best diversity--quality--compute frontier. Anchored regeneration can be strong in final-pool diversity, but its advantage shrinks under full-pipeline token accounting. These results establish practical anchorless baselines for open-ended LLM ideation.

Problem

• LLM ideation
• candidate-pool diversification
• anchorless methods
• semantic diversity
• parallel inference

Innovation

• anchorless diversification
• semantic direction stratification
• parallel LLM ideation
• population-referential divergence
• inference-time control