Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates inference-time controls for diversifying candidate-idea pools generated by LLMs for creative tasks, comparing anchorless methods against anchor-based approaches. They find that population-referential divergence (prompting for ideas different from a reference set) is a strong, low-cost baseline, and semantic direction stratification (planning a set of diverse semantic directions before generation) achieves the best diversity-quality-compute trade-off. The study demonstrates that anchorless methods can effectively rival anchor-dependent methods in open-ended LLM ideation.
Forget prompting from examples - steering LLMs towards diverse ideas is surprisingly effective with simple "anti-example" prompting and strategic planning.
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.