be-creative

AI brainstorming yields generic variations because alignment-trained models produce safe, centroid-biased answers. This Skill solves that by applying verbalized sampling to generate low-probability, high-fit candidates and selecting a non-obvious winner.

To generate distinctive product naming options, this Skill applies tail-driven verbalized sampling, forcing the model to explore diverse, low-probability candidates and output a rationale for the best-fitting winner against your brief.

Verbalized sampling improves creative ideation by generating multiple candidates with explicit probability estimates targeted at ≤ 0.10. It selects the best-fitting option from the tail, preventing repetitive, centroid-biased outputs.

Yes, you can use this approach for story premises, metaphor discovery, taglines, and headline writing. It is designed for any creative task with many valid solutions and meaningful novelty constraints.

To pick a non-obvious winner, ask the model for five diverse options with probability-aware selection rationale, then instruct it to apply verbalized sampling and pick the best non-obvious winner against your brief.

You should avoid tail-driven sampling for tasks requiring deterministic, single correct answers. It is specifically built for divergent ideation where many valid solutions exist and novelty is a meaningful constraint.