Search

CN-122029531-A - Systems and methods for hybrid artificial intelligence enhancement and optimization

CN122029531ACN 122029531 ACN122029531 ACN 122029531ACN-122029531-A

Abstract

Examples are disclosed for validating and optimizing responses generated by a Large Language Model (LLM). A method of multitube concurrency includes implementing a logic programming engine that parses a response into verifiable logic predicates to detect contradictions with a real-world knowledge base. Responses may also be weighted for category verification by modeling them as mathematical morphology shots using category theory to identify semantic inconsistencies that may indicate spurious information associated with the responses. These responses may be modified and supplemented in view of the verification operations.

Inventors

  • X.Ge

Assignees

  • X·葛

Dates

Publication Date
20260512
Application Date
20240919
Priority Date
20230919

Claims (8)

  1. 1. A method for enhancing responses from a Large Language Model (LLM), comprising: (a) Accessing at least one response to the query from the large language model; (b) Parsing the at least one response into a logic programming predicate defining a plurality of facts and actions associated with the at least one response; (c) In view of the knowledge base for logic programming, performing an accuracy (correctness) verification operation for the at least one response, comprising: (c) (i) verifying and retaining facts of the plurality of facts in the at least one response that are indicated as true via the knowledge base, and (C) (ii) removing any other facts of the plurality of facts from the at least one response, the any other facts defining negations that are indicated as true in the knowledge base, and (D) Based on the accuracy verification operation, an output is generated that evaluates the at least one response and returns a verified version of the at least one response.
  2. 2. The method of claim 1, further comprising: determining that the knowledge base cannot verify the plurality of facts, and Performing category verification, including: generating one or more hints that are changed from the query but are semantically equivalent to the query, For each of the one or more hints, obtaining a respective response from the large language model, Establishing a structural relationship between the generated hints and corresponding responses from the large language model, an Any semantic inconsistencies are detected using domain theory, wherein the identification of semantic inconsistencies reflects a lack of reliability associated with the at least one response.
  3. 3. The method of claim 1, wherein the logical programming predicate is defined from logical programming sentences associated with the at least one response and includes conjunctions to construct relationships between the logical programming sentences.
  4. 4. The method of claim 1, further comprising: A multi-agent optimization solution is performed to supplement the at least one response with additional functionality or information associated with the query.
  5. 5. The method of claim 4, further comprising: The large language model is iterated to facilitate formulating at least one mathematical model to customize a modified version of the response based at least in part on the query.
  6. 6. The method of claim 4, further comprising supplementing the at least one response by: generating a plurality of agents, each of the plurality of agents configured to solve a respective optimization problem associated with the query using linear programming, dynamic programming, or optimal control, and Each agent is assigned an objective function and constraints.
  7. 7. The method of claim 6, wherein each agent of the plurality of agents exchanges constraints and decisions with other agents in a plurality of iterations under direction of the LLM until convergence to a globally optimal solution.
  8. 8. The method of claim 1, further comprising: parsing the at least one response into the logical programming predicate to construct an input message for an enterprise computer system; Integrating the input message with an existing enterprise computer system by formatting the constructed input message as an Application Programming Interface (API) call for communication with the existing enterprise computer system, the existing enterprise computer system including an information system and a control system, and Distributing the validated version of the at least one response to the existing enterprise computer system via an API enables automated data query and control actions based on the generated artificial intelligence output.

Description

Systems and methods for hybrid artificial intelligence enhancement and optimization Cross reference to related applications This is a PCT patent application, the present application claims the benefit of U.S. provisional application Ser. No. 63/5391102, filed on 9/19 at 2023, which provisional application is incorporated by reference in its entirety. Technical Field The present disclosure relates generally to Artificial Intelligence (AI) including Large Language Models (LLMs), and more particularly to systems and methods for enhancing the capabilities of conversational AI systems to provide more accurate and reliable responses in enterprise settings across different industries. Background Large language models such as ChatGPT may sometimes produce false or phantom responses that are not suitable for enterprise applications. These models produce answers probabilistically through vast statistical calculations, providing human-like insight. However, the probabilistic-based nature of these models results in inadequate accuracy of their output. Furthermore, these models may program false information without a mechanism for maintaining authenticity, and enhancement models themselves often fail to address such illusion problems. It is with respect to these observations, as well as other considerations, that various aspects of the present disclosure have been conceived and developed. Disclosure of Invention The present disclosure provides examples describing systems and methods for enhancing the capabilities of a conversational AI system, such as ChatGPT, to provide more accurate and reliable responses in enterprise settings across different industries. The terms "operable," "configured to," and "capable of" as used herein are interchangeable in the context of the disclosed methods, apparatus, techniques, devices, systems, etc. In general, the innovations described herein aim to enhance the capabilities of conversational AI systems, such as ChatGPT, to provide more accurate and reliable responses in enterprise settings across different industries, including: energy-utilizing utility grid data to provide outage diagnosis and restoration strategies. Renewable energy expansion is planned based on the power generation assets. Finance-making personalized investment advice based on a personal portfolio. The knowledge of the normal pattern is used to detect fraudulent transactions. Traffic—consider factors such as weather, traffic, fleet availability, etc. to suggest an optimized logistics plan. The asset is tracked using a geospatial database. Manufacturing-interfacing with IoT sensors for predictive maintenance. Production is optimally scheduled based on demand forecast and machine availability. Retail—recommend interfacing with IoT sensors for predictive maintenance. Production is optimally scheduled based on demand forecast and machine availability. The innovation is based on logical reasoning, mathematical optimization, and integration with enterprise data systems, which allows it to override general conversational capabilities. It can provide custom-built, validated solutions that are tied to specific operational reality of different enterprises and industries. In a first set of illustrative examples, the innovation can take the form of a method that includes (a) accessing at least one response to a query from a large language model, (b) parsing the at least one response into a logical programming predicate defining a plurality of facts and actions associated with the at least one response, (c) performing an accuracy (correctness) verification operation for the at least one response in view of a knowledge base for logical programming, including (c) (i) verifying and retaining facts from the at least one response that are indicated as true via the knowledge base, and (c) (ii) removing any other facts from the at least one response that are indicated as true negation in the knowledge base, and (d) generating an output evaluating the at least one response and returning a verified version of the at least one response based on the accuracy verification operation. The method may further include the steps of determining that the knowledge base cannot verify the plurality of facts, and performing category verification including generating one or more hints that are altered from the query but that are semantically equivalent to the query, for each of the one or more hints, obtaining respective responses from the large language model, establishing structural relationships between the generated hints and the corresponding responses from the large language model, and detecting any semantic inconsistencies using category theory, wherein identification of the semantic inconsistencies reflects a lack of reliability associated with the at least one response. In a second set of illustrative examples, the innovation can take the form of a system including a processor in communication with one or more computing devices implementing a