With the popularity of generative AI, the prompt word trading market is rapidly expanding. However, the current trading platform represented by PromptBase still uses seller-driven pricing and lacks objective price measurement standards. Faced with this challenge, the Multimedia and Intelligent Security Team of Fudan University proposed an innovative prompt word trading model designed to better adapt to the future buyer's market.
This new transaction model mainly includes two stages: prompt word category selection and pricing strategy formulation. In the first stage, the platform uses a multi-armed bandit algorithm based on greedy search to select categories of prompt words for sale based on quality assessment. In the second stage, the cascading Stackelberg game method is adopted, which considers buyers, platforms and sellers as first-level leaders, second-level leaders and followers respectively, giving priority to the interests of buyers.
The core of this model is to comprehensively consider the relevance and quality of prompt words and generated content, allowing transaction parties to formulate optimal strategies after weighing costs and income. By setting a reasonable price range and prompt word richness requirements, this model effectively balances the interests of the three parties and is expected to bring a win-win situation.
Researchers Meiling Li and Hongrun Ren elaborated on this pattern in a recent paper published on arXiv. They believe that this trading model can not only better standardize the prompt word market, but also potentially reduce the cost of content creators and improve creation efficiency.
As the number of prompt word products increases and transaction costs decrease, this model is expected to reshape the AI content creation ecosystem. However, the research team also pointed out that factors such as the design of the profit function of the parties to the transaction and the quality evaluation of the prompt words are still the key to affecting the final pricing. In the future, they plan to extend this result to a wider range of prompt word pricing scenarios.
This research provides new ideas for solving the prompt word pricing problem and is expected to play an important role in future AI content creation and transactions.
Address: https://arxiv.org/pdf/2405.15154