SkillOpt from MSFT treats skills as trainable parameters (microsoft.github.io)

Microsoft has introduced SkillOpt, a novel approach that treats skills as trainable parameters within AI agents. Unlike traditional methods that require model fine-tuning or manual prompt adjustments, SkillOpt operates by running a frozen agent on scored batches of data. It employs a separate optimization model that suggests structured edits to enhance performance, only accepting candidate edits when there's a measurable improvement in validation outcomes. This innovation holds significant promise for the AI/ML community by streamlining the skill acquisition process for agents, effectively reducing the reliance on extensive model retraining. The method allows for a more dynamic and adaptable system where skills can be optimized in response to performance metrics, making it easier for developers to enhance agent capabilities without extensive resource investment. As SkillOpt automates the optimization of agent skills, it could pave the way for more efficient and effective AI systems capable of evolving in real-time based on user interactions and feedback.