Modulai collaborated with Expressen to enhance their AI-powered headline generation tool. By using previously published articles as few-shot examples, the team developed a data-driven solution that generates headlines and subheadings tailored to each brand's style and tonality, with minimal additional development effort as models improve.
Challenge
Expressen had developed an AI-powered tool to help editors create headlines and subheadings more quickly and to inspire them with fresh ideas. While functional, scaling the system to accommodate different tonalities and brands was a challenge. Expressen sought Modulai's expertise to create a more data-driven solution that would adapt to each brand's unique requirements without relying on numerous prompts.
Solution
Modulai leveraged large language models to generate headlines. By utilizing few-shot learning, the team developed a single prompt that could generate output tailored to each brand's unique style and tonality. This was achieved by providing the model with historical examples of published content from each brand, allowing it to learn and adapt to the desired style without needing multiple prompts.
Tools
The solution is based on large language models with few-shot learning. Historical examples of published content from each brand are provided as in-context examples, enabling a single prompt to adapt to different brand styles and tonalities without additional fine-tuning or prompt engineering.
Value created
The solution proved to be simple yet effective, and it is expected to remain relevant as AI models become more advanced and cost-effective in the future, with minimal additional development effort required.
Learn more
