Embedding by Elicitation: Dynamic Representations for Bayesian Optimization of System Prompts
The paper discusses a new framework called ReElicit for optimizing system prompts in AI using Bayesian methods. It addresses the challenge of tuning prompts based on aggregate feedback rather than detailed critiques. The results indicate that this approach can enhance the performance of AI systems by adapting to feedback over time.
- ▪ReElicit is a Bayesian optimization framework designed for system prompts in AI.
- ▪The framework operates under conditions where feedback is limited to aggregate metrics.
- ▪ReElicit has shown strong performance across multiple optimization tasks, outperforming existing methods.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.19093 (cs) [Submitted on 18 May 2026] Title:Embedding by Elicitation: Dynamic Representations for Bayesian Optimization of System Prompts Authors:Zhiyuan Jerry Lin, Benjamin Letham, Samuel Dooley, Maximilian Balandat, Eytan Bakshy View a PDF of the paper titled Embedding by Elicitation: Dynamic Representations for Bayesian Optimization of System Prompts, by Zhiyuan Jerry Lin and 4 other authors View PDF HTML (experimental) Abstract:System prompts are a central control mechanism in modern AI systems, shaping behavior across conversations, tasks, and user populations. Yet they are difficult to tune when feedback is available only as aggregate metrics rather than per-example labels, failures, or critiques.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
