Production conversational AI often suffers from "persona drift" where the AI's voice becomes inconsistent over time. The author argues this is an architectural issue, not a prompt problem, suggesting hard-coded logic for character and rules, and introduces ArcOS, a five-stage engine to manage this.
This paper introduces ReElicit, a Bayesian optimization framework based on "embedding by elicitation" for tuning system prompts in AI. It leverages LLMs to elicit an interpretable feature space and a Gaussian process surrogate to select and refine prompts based on aggregate feedback.
Este artigo explora a transição de bibliotecas de prompts trabalhosas para o uso de system prompts, que revolucionaram a maneira como se interage com IAs. System prompts são instruções fundamentais que moldam as respostas da IA, permitindo uma comunicação mais personalizada e contextualizada.