Prompting Techniques and Templates

New prompting techniques include Verbalized Sampling and a compact editorial HTML prompt template.
Key insights
Verbalized Sampling prompts models to enumerate answer probabilities
Stanford’s Verbalized Sampling technique adds roughly twenty words to prompts that instruct the model to generate multiple possible answers and associated probabilities. Reported results include creativity increases of 1.6–2×, a 25.7% rise in diversity, and recovery of 66.8% of creativity reduced by alignment, without requiring retraining and outperforming fine-tuned models. The method addresses mode collapse and promotes diverse model expression. (Post 15609)
Compact editorial prompt template for dense mobile info cards
A ready-to-use prompt and style specification was published for producing single-file HTML5 magazine-style info cards optimized for mobile feeds. The template prescribes Compact Editorial layout, specific color palette, fonts ('Noto Serif SC', 'Oswald', 'Noto Sans SC'), Tailwind CDN inclusion, SVG noise texture, tight grids, and an emphasized dark stats footer. Output instructions require complete HTML code only. (Post 15630)