DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-8, 10-17, 19 and 20 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Aly et al. (U.S. Publication No. 2026/0050771 A1, hereinafter referred to as “Aly”).
Regarding claim 1, Aly discloses a computer-implemented method comprising: ()(e.g., abstract, figures 6, 8 and 9 and paragraph [0060])
receiving a prompt instructing a large language model to generate a target response; (“For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112.” “non-limiting method 600 can comprise employing LLM 112 (e.g., by task determination component 204) to create a prompt given question 122, the set of algorithms 120 and the natural language descriptions of respective algorithms in the set of algorithms 120. The prompt can also comprise questions that are semantically similar to question 122 and responses of such questions, and the prompt can be employable to determine the one or more tasks that can be executed to generate a response to question 122.”)(e.g., paragraphs [0061] and [0062])
identifying, from among a repository of retrieval-augmented-thought (RAT) store items comprising text descriptions interpretable by the large language model to inform chain-of-thought response generation, a RAT-store item corresponding to the prompt; (“Recall that agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and their corresponding solutions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. A positive feedback on response 124 can trigger an update to the vector database, reinforcing the association between question 122 and a successful resolution to question 122. For example, access component 202 can access the positive feedback and task determination component 204 can update the vector database. Conversely, negative feedback can trigger agent 110 to de-prioritize question 122 and response 124 as less effective examples, which in some embodiments, can lead to the generation of new embeddings that can more accurately capture the intent of a question. Such a continuous learning loop can ensure that the agent 110 and system 102 evolve over time, thereby delivering increasingly accurate and contextually relevant responses. In this regard, agent 110 can be a machine learning model that can employ chain-of-thought reasoning to generate responses to questions, which can continuously improve the accuracies of the responses.”)(e.g., paragraphs [0032] and [0045])
retrieving the RAT-store item from the repository of RAT-store items; and (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. Upon accessing a new question (or query), agent 110 can search the vector database for semantically similar questions. The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “Thus, the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0032] and [0053])
generating a response by utilizing the large language model to execute a sequence of processes indicated by the RAT-store item. (“The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112. Each algorithm in the set of algorithms 120 can be associated with a natural language description (e.g., algorithm 1—description, algorithm 2—description, etc.). The respective natural language descriptions of the respective algorithms can assist LLM 112 to determine a task that an algorithm can execute. A set of exemplary questions that can be semantically similar to question 122 can be additionally provided to LLM 112 via the prompt, based on which LLM 112 can determine one or more tasks and the corresponding algorithms executable to solve complex analytical problems. For example, agent 110 can employ a chain-of-thought reasoning technique (prompt technique) in conjunction with LLM 112 to divide question 122 into a series of tasks/steps that can be executable via algorithms comprised in the set of algorithms 120.”)(e.g., figures 6, 8 and 9 and paragraphs [0032] and [0061])
Regarding claim 2, Aly discloses the computer-implemented method of claim 1. Aly further discloses further comprising generating the repository of RAT-store items by generating a plurality of RAT-store items, wherein: (“The existing technique can solve financial problems by employing an LLM to draw inference from a large corpus of financial text information. Additionally, the existing technique is directed to explainability rather than reasoning. On the contrary, the various embodiments of the present disclosure can employ an LLM and a sequence of algorithms to solve complex business intelligence or analytics based questions, wherein the output of an algorithm can direct the sequence and influence the selection of algorithms in the sequence.”)(e.g., paragraph [0022])
a first subset of the RAT-store items include text descriptions guiding execution of the sequence of processes by the large language model for generating the response; and (“Additionally, respective algorithms comprised in the set of algorithms 120 can be associated with respective names and respective natural language descriptions, wherein the natural language description of an algorithm can describe the type of task that the algorithm can execute. In various embodiments, task determination component 204 can input the prompt into LLM 112. LLM 112 can process the prompt and determine the one or more tasks executable to generate response 124. For example, the questions retrieved from the vector database can indicate the type of questions that LLM 112 can expect to encounter in the future. For example, based on the questions retrieved from the vector database, LLM 112 can determine the type or category of question 122. Further, the questions retrieved from the vector database can be questions previously processed by LLM 112, and LLM 112 can access (e.g., from a memory accessible to LLM 112) historical knowledge about the tasks and the corresponding algorithms previously executed by LLM 112 to generate the responses to the questions retrieved from the vector database.”)(e.g., paragraphs [0022], [0035] and [0104])
a second subset of the RAT-store items include text descriptions of example data interpretable by the large language model to execute the sequence of processes for generating the response. (“Based on the historical knowledge and the semantic similarity of the questions retrieved from the vector database to question 122, LLM 112 can identify the one or more tasks that can be executed to generate response 124 for question 122. Further, LLM 112 can identify a sequence in which the one or more tasks can be executed. It should be appreciated that question 122 can be a query or a sentence, such as a request for statistical or other type of information.”)(e.g., paragraphs [0022], [0035] and [0104]).
Regarding claim 3, Aly discloses the computer-implemented method of claim 1. Aly further discloses further comprising generating a hybrid RAT-store item from a first RAT-store item and a second RAT-store item from among the repository of RAT-store items by combining at least one text description from the first RAT-store item and at least one text description from the second RAT-store item into the hybrid RAT-store item. (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. Upon accessing a new question (or query), agent 110 can search the vector database for semantically similar questions. The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.”)(e.g., paragraphs [0032]-[0035]).
Regarding claim 4, Aly discloses the computer-implemented method of claim 1. Aly further discloses further comprising: receiving feedback data from a client device based on the response generated by the large language model; and (“feedback component 402 can provide a feedback mechanism employable to generate feedback on response 124. For example, feedback component 402 can provide an interactive feedback mechanism, wherein entities (e.g., hardware, software, machine, AI, neural network and/or users) can select a “thumbs-up” or “thumbs-down” option for response 124. The “thumbs-up” option can indicate a positive or favorable feedback, and the “thumbs-down” option can indicate a negative or unfavorable feedback. Entities can select such options at a UI of a device via the click of a mouse button, a touchscreen mechanism, voice, or another suitable mechanism. Recall that agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and their corresponding solutions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. A positive feedback on response 124 can trigger an update to the vector database, reinforcing the association between question 122 and a successful resolution to question 122. For example, access component 202 can access the positive feedback and task determination component 204 can update the vector database. Conversely, negative feedback can trigger agent 110 to de-prioritize question 122 and response 124 as less effective examples, which in some embodiments, can lead to the generation of new embeddings that can more accurately capture the intent of a question.”)(e.g., paragraph [0045])
based on the feedback data, updating parameters of the large language model to modify how the large language model identifies relevant RAT-store items for received prompts. (“Conversely, negative feedback can trigger agent 110 to de-prioritize question 122 and response 124 as less effective examples, which in some embodiments, can lead to the generation of new embeddings that can more accurately capture the intent of a question. Such a continuous learning loop can ensure that the agent 110 and system 102 evolve over time, thereby delivering increasingly accurate and contextually relevant responses. In this regard, agent 110 can be a machine learning model that can employ chain-of-thought reasoning to generate responses to questions, which can continuously improve the accuracies of the responses.”)(e.g., paragraph [0045]).
Regarding claim 5, Aly discloses the computer-implemented method of claim 1. Aly further discloses further comprising generating a new RAT-store item to include within the repository of RAT-store items based on: (“Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., figures 6, 8 and 9 and paragraph [0053])
determining that the repository of RAT-store items does not include a relevant RAT-store item for the prompt; and (“Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response.”)(e.g., figures 6, 8 and 9 and paragraph [0053])
generating text describing processes executable by the large language model to generate the target response by using the large language model to model the text after existing RAT-store items within the repository of RAT-store items. (“Agent 110 can employ LLM 112 and a chain-of-thought reasoning to determine/create one or more tasks that can be executed to generate a response to question 122, wherein each task can be associated with a reasoning. Agent 110 can further select and execute suitable algorithms to execute the one or more tasks. Agent 110 can engage an algorithm to execute a task of the one or more tasks, and upon execution of the task, agent 110 can parse and validate/verify (via LLM 112) the output of the task. If the output represents an invalid or incomplete response to question 122, agent 110 can determine, via LLM 112, a subsequent task to be executed, based on the output of a previously executed task. The process of determining a task to be executed, selecting and executing an algorithm to execute the task, parsing the output of the task and validating the output can continue until agent 110 can validate that an output represents a response (e.g., response 124) to question 122.”)(e.g., figures 6, 8 and 9 and paragraph [0053]-[0056]).
Regarding claim 6, Aly discloses the computer-implemented method of claim 1. Aly further discloses wherein the RAT-store item comprises a stored content item that includes text descriptions of the sequence of processes that, when interpreted by the large language model, instructs the large language model to execute the sequence of processes to generate the target response. (“the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0040] and [0053])
Regarding claim 7, Aly discloses the computer-implemented method of claim 1. Aly further discloses further comprising: determining, from the RAT-store item, a first RAT process to execute from among the sequence of processes; (“In various embodiments, task determination component 204 can employ LLM 112 to subdivide the query into multiple tasks, and task execution component 206 can employ LLM 112 to select and execute algorithms to execute the tasks. For example, task determination component 204 can determine, via LLM 112, that the first task to be executed can be the identification of the city with the biggest drop in revenue. Thereafter, task execution component 206 can select and execute, via LLM 112, an algorithm to execute the first task.” “The agent can create or suggest a list of tasks based on the question, the one or more algorithms, and an LLM. The agent can execute a first task of the list of tasks.”)(e.g., paragraphs [0049] and [0101])
upon executing the first RAT process, determining, utilizing a RAT replanner, a second RAT process to execute from among the sequence of processes of the RAT-store item; and (“Further, the agent can parse and validate an output of the first task against the question. Finally, based on a determination that the output of the first task does not answer the question, the agent can employ the output of the first task and execute a second task selected from the list of tasks. The agent can validate the output of the second task and the process of executing additional tasks can conclude upon a determination by the agent that the question has been answered.”)(e.g., paragraphs [0050]-[0055] and [0101])
generating the response by executing the first RAT process and the second RAT process utilizing the large language model. (“The communication from the algorithm to LLM 112 can be in the form of code, and rephrasing component 304 can transform the code into natural language that LLM 112 can comprehend. Thereafter, validation component 302 can validate that the output of the third task indicates that question 122 has been completely answered. Thus, in various embodiments, each algorithm can generate a premature output, and agent 110 can determine whether the output represents a complete response (e.g., response 124) to question 122. If so, the output can be presented to an end entity, for example, an entity that generated question 122.” “the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0050]-[0055] and [0101])
Regarding claim 8, Aly discloses the computer-implemented method of claim 7. Aly further discloses further comprising determining, upon executing the first RAT process, replanner data by utilizing the RAT replanner to determine contextual data for informing execution of the second RAT process by the large language model after execution of the first RAT process. (“Further, task execution component 206 can select and execute, via LLM 112, a different algorithm from the set of algorithms to execute the new task. In another embodiment, task execution component 206 can execute the subsequent task with more context or a different algorithm, upon a determination that new output 208 represents an invalid, incomplete or failed response to question 122. In yet another embodiment, rephrasing component 304 can generate response 124 by transforming (e.g., translating) new output 208 to a format applicable to question 122, upon a determination that new output 208 represents a complete and valid response to question 122.”)(e.g., paragraphs [0039], [0045], [0057] and [0061]).
Regarding claim 10, Aly discloses a system comprising: at least one processor; and at least one non-transitory computer-readable storage medium storing instructions that, when executed by the at least one processor, cause the system to: (e.g., paragraph [0027])
receive a prompt instructing a large language model to generate a target response; (“For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112.” “non-limiting method 600 can comprise employing LLM 112 (e.g., by task determination component 204) to create a prompt given question 122, the set of algorithms 120 and the natural language descriptions of respective algorithms in the set of algorithms 120. The prompt can also comprise questions that are semantically similar to question 122 and responses of such questions, and the prompt can be employable to determine the one or more tasks that can be executed to generate a response to question 122.”)(e.g., paragraphs [0061] and [0062])
identify, from among a repository of RAT-store items, a RAT-store item corresponding to the prompt, wherein the RAT-store item comprises a sequential text description of a sequence of processes executable by the large language model; (“Recall that agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and their corresponding solutions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. A positive feedback on response 124 can trigger an update to the vector database, reinforcing the association between question 122 and a successful resolution to question 122. For example, access component 202 can access the positive feedback and task determination component 204 can update the vector database. Conversely, negative feedback can trigger agent 110 to de-prioritize question 122 and response 124 as less effective examples, which in some embodiments, can lead to the generation of new embeddings that can more accurately capture the intent of a question. Such a continuous learning loop can ensure that the agent 110 and system 102 evolve over time, thereby delivering increasingly accurate and contextually relevant responses. In this regard, agent 110 can be a machine learning model that can employ chain-of-thought reasoning to generate responses to questions, which can continuously improve the accuracies of the responses.”)(e.g., paragraphs [0032] and [0045])
retrieve the RAT-store item from the repository of RAT-store items; and (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. Upon accessing a new question (or query), agent 110 can search the vector database for semantically similar questions. The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “Thus, the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0032] and [0053])
generate a response by utilizing the large language model to execute the sequence of processes defined by the RAT-store item. (“The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112. Each algorithm in the set of algorithms 120 can be associated with a natural language description (e.g., algorithm 1—description, algorithm 2—description, etc.). The respective natural language descriptions of the respective algorithms can assist LLM 112 to determine a task that an algorithm can execute. A set of exemplary questions that can be semantically similar to question 122 can be additionally provided to LLM 112 via the prompt, based on which LLM 112 can determine one or more tasks and the corresponding algorithms executable to solve complex analytical problems. For example, agent 110 can employ a chain-of-thought reasoning technique (prompt technique) in conjunction with LLM 112 to divide question 122 into a series of tasks/steps that can be executable via algorithms comprised in the set of algorithms 120.”)(e.g., figures 6, 8 and 9 and paragraphs [0032] and [0061])
Regarding claim 11, Aly discloses the system of claim 10. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the system to identify the RAT-store item corresponding to the prompt by: extracting a prompt embedding from the prompt; and (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs.”)(e.g., paragraphs [0032]-[0034])
comparing the prompt embedding to RAT-store item embeddings extracted from the RAT-store items to determine a relevant RAT-store item. (“For example, task determination component 204 can search a vector database comprising embeddings of questions and embeddings of responses to the questions.” “The embeddings thus generated can capture the semantic meaning of the text in question 122 and the questions stored in the vector database, thereby allowing task determination component 204 to measure the similarity between questions beyond exact keyword matches.”)(e.g., paragraphs [0032]-[0034]).
Regarding claim 12, Aly discloses the system of claim 10. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the system to generate the response by: determining, as informed by the RAT-store item, a content item stored in a content item database accessible by the large language model to analyze as part of the sequence of processes defined by the RAT-store item; and (“the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0040] and [0053])
executing, utilizing the large language model, the sequence of processes by analyzing the content item from the content item database. (“Further, task execution component 206 can orchestrate the execution of the one or more algorithms, via LLM 112, according to the sequence determined by LLM 112 for the one more tasks. Accordingly, task execution component 206 can identify and select/engage, via LLM 112, the algorithm that can execute the first task. For example, to engage an algorithm to execute a task, LLM 112 can input information from question 122 into the algorithm, and only the information from question 122 that can be applicable to the first task can be input into the algorithm.”)(e.g., paragraphs [0037], [0038] and [0073])
Regarding claim 13, Aly discloses the system of claim 10. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the system to generate a hybrid RAT-store item from a first RAT-store item and a second RAT-store item from among the repository of RAT-store items by combining at least one text description from the first RAT-store item and at least one text description from the second RAT-store item into the hybrid RAT-store item. (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. Upon accessing a new question (or query), agent 110 can search the vector database for semantically similar questions. The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.”)(e.g., paragraphs [0032]-[0035]).
Regarding claim 14, Aly discloses the system of claim 10. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the system to: determine, from the RAT-store item, a first RAT process to execute from among the sequence of processes; (“In various embodiments, task determination component 204 can employ LLM 112 to subdivide the query into multiple tasks, and task execution component 206 can employ LLM 112 to select and execute algorithms to execute the tasks. For example, task determination component 204 can determine, via LLM 112, that the first task to be executed can be the identification of the city with the biggest drop in revenue. Thereafter, task execution component 206 can select and execute, via LLM 112, an algorithm to execute the first task.” “The agent can create or suggest a list of tasks based on the question, the one or more algorithms, and an LLM. The agent can execute a first task of the list of tasks.”)(e.g., paragraphs [0049] and [0101])
upon executing the first RAT process, determine, utilizing a RAT replanner, a second RAT process to execute from among the sequence of processes of the RAT-store item; and (“Further, the agent can parse and validate an output of the first task against the question. Finally, based on a determination that the output of the first task does not answer the question, the agent can employ the output of the first task and execute a second task selected from the list of tasks. The agent can validate the output of the second task and the process of executing additional tasks can conclude upon a determination by the agent that the question has been answered.”)(e.g., paragraphs [0050]-[0055] and [0101])
generate the response by executing the first RAT process and the second RAT process utilizing the large language model. (“The communication from the algorithm to LLM 112 can be in the form of code, and rephrasing component 304 can transform the code into natural language that LLM 112 can comprehend. Thereafter, validation component 302 can validate that the output of the third task indicates that question 122 has been completely answered. Thus, in various embodiments, each algorithm can generate a premature output, and agent 110 can determine whether the output represents a complete response (e.g., response 124) to question 122. If so, the output can be presented to an end entity, for example, an entity that generated question 122.” “the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0050]-[0055] and [0101])
Regarding claim 15, Aly discloses the system of claim 14. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the system to determine, upon executing the first RAT process, replanner data by utilizing the RAT replanner to determine contextual data for informing execution of the second RAT process by the large language model after execution of the first RAT process. (“Further, task execution component 206 can select and execute, via LLM 112, a different algorithm from the set of algorithms to execute the new task. In another embodiment, task execution component 206 can execute the subsequent task with more context or a different algorithm, upon a determination that new output 208 represents an invalid, incomplete or failed response to question 122. In yet another embodiment, rephrasing component 304 can generate response 124 by transforming (e.g., translating) new output 208 to a format applicable to question 122, upon a determination that new output 208 represents a complete and valid response to question 122.”)(e.g., paragraphs [0039], [0045], [0057] and [0061]).
Regarding claim 16, Aly discloses a non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computing device to: (e.g., paragraph [0109])
receive a prompt instructing a large language model to generate a target response; (“For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112.” “non-limiting method 600 can comprise employing LLM 112 (e.g., by task determination component 204) to create a prompt given question 122, the set of algorithms 120 and the natural language descriptions of respective algorithms in the set of algorithms 120. The prompt can also comprise questions that are semantically similar to question 122 and responses of such questions, and the prompt can be employable to determine the one or more tasks that can be executed to generate a response to question 122.”)(e.g., paragraphs [0061] and [0062])
determine that a repository of RAT-store items available to the large language model includes a RAT-store item corresponding to the prompt, wherein the RAT-store item comprises a text description of a sequence of processes executable by the large language model; (“Recall that agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and their corresponding solutions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. A positive feedback on response 124 can trigger an update to the vector database, reinforcing the association between question 122 and a successful resolution to question 122. For example, access component 202 can access the positive feedback and task determination component 204 can update the vector database. Conversely, negative feedback can trigger agent 110 to de-prioritize question 122 and response 124 as less effective examples, which in some embodiments, can lead to the generation of new embeddings that can more accurately capture the intent of a question. Such a continuous learning loop can ensure that the agent 110 and system 102 evolve over time, thereby delivering increasingly accurate and contextually relevant responses. In this regard, agent 110 can be a machine learning model that can employ chain-of-thought reasoning to generate responses to questions, which can continuously improve the accuracies of the responses.”)(e.g., paragraphs [0032] and [0045])
retrieve the RAT-store item from the repository of RAT-store items; and (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs. Upon accessing a new question (or query), agent 110 can search the vector database for semantically similar questions. The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “Thus, the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0032] and [0053])
generate a response by utilizing the large language model to execute the sequence of processes defined by the RAT-store item. (“The questions retrieved by agent 110, particularly questions with successful outcomes, can be injected into a prompt as examples to guide LLM 112 in generating an optimal chain of thought.” “For example, a prompt comprising the set of algorithms 120, question 122, and contextual information associated with question 122 can be provided to agent 110 or LLM 112. Each algorithm in the set of algorithms 120 can be associated with a natural language description (e.g., algorithm 1—description, algorithm 2—description, etc.). The respective natural language descriptions of the respective algorithms can assist LLM 112 to determine a task that an algorithm can execute. A set of exemplary questions that can be semantically similar to question 122 can be additionally provided to LLM 112 via the prompt, based on which LLM 112 can determine one or more tasks and the corresponding algorithms executable to solve complex analytical problems. For example, agent 110 can employ a chain-of-thought reasoning technique (prompt technique) in conjunction with LLM 112 to divide question 122 into a series of tasks/steps that can be executable via algorithms comprised in the set of algorithms 120.”)(e.g., figures 6, 8 and 9 and paragraphs [0032] and [0061]).
Regarding claim 17, Aly discloses the non-transitory computer-readable medium of claim 16. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the computing device to identify the RAT-store item corresponding to the prompt by determining, using the large language model, a relevant RAT-store item that includes text descriptions of the sequence of processes that are executable by the large language model to generate the target response indicated by the prompt. (“agent 110 can utilize a vector database to store and retrieve embeddings of questions previously addressed by agent 110 and the solutions corresponding to such questions, to enhance the accuracy and relevance of the chain-of-thought reasoning employed by agent 110 in processing natural language inputs.” “For example, task determination component 204 can search a vector database comprising embeddings of questions and embeddings of responses to the questions.” “The embeddings thus generated can capture the semantic meaning of the text in question 122 and the questions stored in the vector database, thereby allowing task determination component 204 to measure the similarity between questions beyond exact keyword matches.”)(e.g., paragraphs [0032]-[0034]).
Regarding claim 19, Aly discloses the non-transitory computer-readable medium of claim 16. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the computing device to: determine, from the RAT-store item, a first RAT process to execute from among the sequence of processes; (“In various embodiments, task determination component 204 can employ LLM 112 to subdivide the query into multiple tasks, and task execution component 206 can employ LLM 112 to select and execute algorithms to execute the tasks. For example, task determination component 204 can determine, via LLM 112, that the first task to be executed can be the identification of the city with the biggest drop in revenue. Thereafter, task execution component 206 can select and execute, via LLM 112, an algorithm to execute the first task.” “The agent can create or suggest a list of tasks based on the question, the one or more algorithms, and an LLM. The agent can execute a first task of the list of tasks.”)(e.g., paragraphs [0049] and [0101])
upon executing the first RAT process, determine, utilizing a RAT replanner, a second RAT process to execute from among the sequence of processes of the RAT-store item; and (“Further, the agent can parse and validate an output of the first task against the question. Finally, based on a determination that the output of the first task does not answer the question, the agent can employ the output of the first task and execute a second task selected from the list of tasks. The agent can validate the output of the second task and the process of executing additional tasks can conclude upon a determination by the agent that the question has been answered.”)(e.g., paragraphs [0050]-[0055] and [0101])
generate the response by executing the first RAT process and the second RAT process utilizing the large language model. (“The communication from the algorithm to LLM 112 can be in the form of code, and rephrasing component 304 can transform the code into natural language that LLM 112 can comprehend. Thereafter, validation component 302 can validate that the output of the third task indicates that question 122 has been completely answered. Thus, in various embodiments, each algorithm can generate a premature output, and agent 110 can determine whether the output represents a complete response (e.g., response 124) to question 122. If so, the output can be presented to an end entity, for example, an entity that generated question 122.” “the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0050]-[0055] and [0101])
Regarding claim 20, Aly discloses the non-transitory computer-readable medium of claim 19. Aly further discloses further comprising instructions that, when executed by the at least one processor, cause the computing device to determine, upon executing the first RAT process, replanner data by utilizing the RAT replanner to determine contextual data for informing execution of the second RAT process by the large language model after execution of the first RAT process. (“Further, task execution component 206 can select and execute, via LLM 112, a different algorithm from the set of algorithms to execute the new task. In another embodiment, task execution component 206 can execute the subsequent task with more context or a different algorithm, upon a determination that new output 208 represents an invalid, incomplete or failed response to question 122. In yet another embodiment, rephrasing component 304 can generate response 124 by transforming (e.g., translating) new output 208 to a format applicable to question 122, upon a determination that new output 208 represents a complete and valid response to question 122.”)(e.g., paragraphs [0039], [0045], [0057] and [0061]).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Aly in view of Barrett et al. (U.S. Publication No. 2024/0346096 A1, hereinafter referred to as “Barrett”).
Regarding claim 9, Aly discloses the computer-implemented method of claim 7. Aly further discloses wherein generating the response comprises: determining, according to the RAT-store item, a function adapter from among a plurality of candidate function adapters, the function adapter comprising computer code executable to perform the first RAT process indicated by a text description of the sequence of processes in the RAT-store item; and (“the various embodiments herein can provide a system that can execute a chain-of-thought reasoning process, wherein the system can constantly evaluate whether a question has been resolved. If not, the system can determine the subsequent tasks to be executed to resolve the question. Additionally, the system can comprise intelligence that can recognize whether the solution for a task has been attained. Accordingly, the system can redirect the output of the task to a subsequent task in the process, instead of repeating the task. For example, if the output of a task represents an invalid or partially complete response to the question, the system can execute additional tasks to generate a complete response. In various embodiments, the system can maintain the questions that have been successfully resolved in a memory.”)(e.g., paragraphs [0040] and [0053])
However, Aly does not appear to specifically disclose identifying, utilizing the function adapter to perform the first RAT process, genealogical information corresponding to the first RAT process from a genealogical database associated with a genealogical-data system.
On the other hand, Barrett, which relates to a personal helper bot system with example applications in genealogy (title) does disclose identifying, utilizing the function adapter to perform the first RAT process, genealogical information corresponding to the first RAT process from a genealogical database associated with a genealogical-data system. (“The system prowess, exemplified through the performance of genealogical searches and the curation of music playlists, involves streamlining tasks traditionally requiring extensive manual searches across various databases and/or the use of multiple applications. By incorporating case-based reasoning machine learning AI algorithms, the system learns from BOT collaborations, thereby improving its performance and reusability over time.”)(e.g., abstract and paragraphs [0074] and [0242]).
Aly discloses response generation based on chain-of-thought reasoning. In Aly, a task determination component determines one or more tasks to be executed to generate a response to a question. E.g., abstract. However, Aly does not appear to specifically disclose that the process involves searching a genealogical database. On the other hand, Barrett, which also employs reasoning based searching (abstract) does disclose that it is beneficial to search genealogy based on previous BOT interactions. Abstract. This provides for streamlined and optimized search efficiency by harnessing the reasoning of previous search interactions. E.g., paragraph [0005]. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s claimed invention to incorporate the genealogy searches as disclosed in Barrett to further enhance the system of Aly to provide users with the enhanced benefit to effectively search data relating to genealogy as described in Barrett.
Regarding claim 18, Aly discloses the non-transitory computer-readable medium of claim 16. However, Aly does not appear to specifically disclose further comprising instructions that, when executed by the at least one processor, cause the computing device to receive the prompt by receiving, from a client device, a text description of an instruction to search a genealogical database to generate the response from genealogical information.
On the other hand, Barrett, which relates to a personal helper bot system with example applications in genealogy (title) does disclose further comprising instructions that, when executed by the at least one processor, cause the computing device to receive the prompt by receiving, from a client device, a text description of an instruction to search a genealogical database to generate the response from genealogical information. (“The system prowess, exemplified through the performance of genealogical searches and the curation of music playlists, involves streamlining tasks traditionally requiring extensive manual searches across various databases and/or the use of multiple applications. By incorporating case-based reasoning machine learning AI algorithms, the system learns from BOT collaborations, thereby improving its performance and reusability over time.”)(e.g., abstract and paragraphs [0074] and [0242]).
It would have been obvious to combine Barrett with Aly for the same reasons as provided in claim 9, above.
Conclusion
The prior art made of record, listed on form PTO-892, and not relied upon is considered pertinent to applicant's disclosure..
Any inquiry concerning this communication or earlier communications from the examiner should be directed to RICHARD L BOWEN whose telephone number is (571)270-5982. The examiner can normally be reached Monday through Friday 7:30AM - 4:00PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Aleksandr Kerzhner can be reached at (571)270-1760. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/RICHARD L BOWEN/ Primary Examiner, Art Unit 2165
Prosecution Insights