Preferred Networks (PFN) has unveiled an innovative feature for its generative AI platform, PreferredAI Work Suite, which automatically optimizes prompts based on qualitative user feedback. In traditional workflows involving complex, interconnected LLM blocks, human operators previously had to manually fine-tune prompts to achieve desired outputs. This new functionality aims to significantly alleviate the burden of prompt engineering by automating these iterative adjustments through a more structured, algorithmic approach.

At its technical core lies an internal framework built upon Optuna, the widely used open-source hyperparameter optimization tool developed by PFN. This framework translates vague user instructions—such as a request to "only output keywords"—into automated prompt updates using a methodology known as Text Gradient. Notably, PFN has introduced control mechanisms specifically designed for environments where multiple LLM blocks coexist. By assigning internal identifiers to each prompt, the system can determine which specific blocks require updates, thereby preventing a single piece of feedback from negatively impacting the performance of other stable components.

The optimization process is visualized in a tree format, allowing users to intuitively accept or reject suggested prompt variations. By leveraging Optuna’s history management capabilities, the system accumulates past failures as learning data to reach the ideal prompt configuration more efficiently. This initiative represents a prime example of PFN applying its long-standing expertise in AutoML to the generative AI domain. The company's ability to rapidly integrate research outcomes into practical products marks a significant step toward accelerating the democratization of AI technology and its practical application in enterprise settings.

Preferred Networks (PFN) has unveiled an innovative feature for its generative AI platform, PreferredAI Work Suite, which automatically optimizes prompts based on qualitative user feedback. In traditional workflows involving complex, interconnected LLM blocks, human operators previously had to manually fine-tune prompts to achieve desired outputs. This new functionality aims to significantly alleviate the burden of prompt engineering by automating these iterative adjustments through a more structured, algorithmic approach.

At its technical core lies an internal framework built upon Optuna, the widely used open-source hyperparameter optimization tool developed by PFN. This framework translates vague user instructions—such as a request to "only output keywords"—into automated prompt updates using a methodology known as Text Gradient. Notably, PFN has introduced control mechanisms specifically designed for environments where multiple LLM blocks coexist. By assigning internal identifiers to each prompt, the system can determine which specific blocks require updates, thereby preventing a single piece of feedback from negatively impacting the performance of other stable components.

The optimization process is visualized in a tree format, allowing users to intuitively accept or reject suggested prompt variations. By leveraging Optuna’s history management capabilities, the system accumulates past failures as learning data to reach the ideal prompt configuration more efficiently. This initiative represents a prime example of PFN applying its long-standing expertise in AutoML to the generative AI domain. The company's ability to rapidly integrate research outcomes into practical products marks a significant step toward accelerating the democratization of AI technology and its practical application in enterprise settings.