Generative Workflow Filling for Adaptive Prompt Chaining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prompt chaining techniques in scaffolded ML systems are limited by their inability to adapt to different input properties or execution conditions, requiring explicit specification of input and output types, and are unable to generate recursive structures or handle multimodal inputs and outputs effectively.
Innovation Solution
A computing system that utilizes a generative language model to dynamically select input sources and output destinations, allowing for the generation of computational workflows with cyclic graph topologies and flexible input-output structures, including retrieval-augmented generation to incorporate additional information and iterative user revisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If prompt chaining techniques are used in scaffolded ML systems, then the system can perform specified tasks using ML models, but the system cannot adapt to different input properties or execution conditions and requires explicit specification of input and output types
Solution Approach 1:
The patent implements dynamic workflow generation where the generative language model automatically adapts input sources and output destinations based on execution conditions. The workflow stages can dynamically select from multiple input sources and destinations without requiring explicit pre-specification, enabling the system to respond to varying real-world conditions while maintaining manageable complexity through automated generation.
2Adaptability or versatility
If prompt chaining techniques are used, then the system can execute computational workflows, but the system cannot generate recursive structures or handle multimodal inputs and outputs effectively
Solution Approach 1:
The patent creates a universal workflow generation framework where a single generative language model handles multiple functions: generating recursive structures, selecting from various input modalities (text, images, audio), and determining output destinations. This multi-functional approach consolidates what would otherwise require separate specialized systems into one unified model, making the system both versatile and easy to create.
3Reliability
If explicit specification of input and output types is required, then the system can maintain structured workflows, but the system loses flexibility in handling real-world variations
Solution Approach 1:
The generative language model performs self-service by automatically determining input sources and output destinations without requiring explicit user specification. The model monitors execution conditions and autonomously selects appropriate workflow stages and data flows, maintaining structural integrity through generated workflows while adapting to real-world variations through its inherent flexibility and learning capabilities.
Data Source
AI summary
A computing system including one or more processing devices configured to receive a computational workflow specification. The computational workflow specification includes a plurality of workflow stages. The plurality of workflow stages include one or more unfilled workflow stages that each include a respective workflow stage objective, a respective workflow stage exit criterion, and one or more fillable fields. The computational workflow specification further includes a directed graph structure in which the plurality of workflow stages are arranged. Based at least in part on the computational workflow specification, the one or more processing devices are further configured to compute a filled computational workflow including one or more filled workflow stages. Computing the filled computational workflow includes computing respective filled values of the one or more fillable fields at least in part at a generative language model. The one or more processing devices are further configured to execute the filled computational workflow.