Microsoft patents system that automatically improves AI prompts
Microsoft describes a method that automatically improves the instructions given to AI models by testing many variations and keeping the best ones. It also searches a database of past examples to add useful context to the final instruction. This helps the AI produce better answers without a human manually tweaking the prompt.
In the filing’s own words
Techniques for generating optimized input prompts for generative language models combines automated prompt refinement with intelligent context retrieval. A prompt optimization service automatically refines initial prompt instructions through controlled mutation and evaluation cycles using multiple generative language models. A context management service maintains a vector database of historical examples pre-encoded as embedding vectors, enabling efficient retrieval of semantically similar examples for new tasks. A dynamic prompt generator assembles optimized prompts by combining refined instructions with relevant historical examples and task context. The system supports deployment with both large language models and smaller fine-tuned models, enabling efficient operation across diverse applications. Configurable parameters allow balancing optimization quality against computational costs. The system maintains efficiency through single-call mutations, fast vector similarity search, and support for compact fine-tuned models that reduce resource requirements while preserving response quality.
What exactly they claim (claim 1)
1. A computer-implemented method for generating an input prompt for a generative language model, the method comprising: receiving task context data for a task and an initial prompt instruction for instructing the generative language model to perform the task; generating an optimized prompt instruction by iteratively applying, with a prompt optimization system that uses one or more generative language models, a plurality of mutation operations and a plurality of refinement operations to the initial prompt instruction, wherein each mutation operation comprises: generating, by a mutator language model, a plurality of candidate prompt variations from a current prompt instruction; evaluating, by a critic language model, the plurality of candidate prompt variations against a set of training samples to assign a performance-based score to each candidate prompt variation; and selecting a highest-scoring candidate prompt variation from the plurality of candidate prompt variations based on the assigned performance-based scores; wherein each refinement operation comprises: generating, by the critic language model, targeted feedback identifying strengths and weaknesses of the highest-scoring candidate prompt variation; modifying the highest-scoring candidate prompt variation based on the targeted feedback to produce a refined prompt instruction; and using the refined prompt instruction as the current prompt instruction in a subsequent mutation operation until reaching a convergence criterion; assembling the input prompt based on the optimized prompt instruction and the task context data; and providing the input prompt to the generative language model to generate a response for performing the task.
Facts
- Company
- Microsoft (Microsoft Technology Licensing, LLC)
- Official title
- AUTOMATED PROMPT OPTIMIZATION FOR GENERATIVE LANGUAGE MODELS
- Publication number
- US20260277977A1
- Application number
- 19680435
- Filed
- May 18, 2026
- Published
- Sep 17, 2026
- Inventors
- 8
- Claims
- 20
Classified under
G06F 16/3347 Physics & computingG06F 11/3604 Physics & computingG06F 16/3322 Physics & computingG16H 50/20 Physics & computingG16H 50/70 Physics & computingMore from Microsoft in AI & software
- Microsoft patents method to check AI answers against verified facts Oct 20, 2026
- Microsoft patents AI that presents PowerPoint slides for you Sep 3, 2026
- Microsoft patents method to check if LLM answers are based on real data Sep 10, 2026
- Microsoft patents system to blend multiple AI agents into one chat Oct 1, 2026
- Microsoft patents system that answers questions using your screen history Oct 15, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.