Prompt engineering computer, prompt engineering method and program

Abstract

To improve the accuracy of answers to questions. [Solution] A prompt engineering computer that uses a generation AI to output answers to prompts acquires the prompt, determines two or more bots to answer the prompt, generates question data for the bots, asks each of the bots questions based on the question data, acquires each answer generated by each of the bots, generates an answer document summarizing each answer as an answer to the prompt, and outputs the answer document as an answer to the prompt.

Description

Technical Field

[0001] The present invention relates to a technology effective for the utilization of generative AI (Artificial Intelligence).

Background Art

[0002] In recent years, the spread of generative AI has been progressing. In generative AI, appropriate prompt engineering is important, and by using appropriate prompts (questions, explanations, instructions, summaries, etc.), the answer accuracy is improved. Currently, there is a limit set on the number of characters of the prompts used for generative AI. For example, in a certain generative AI, this character limit is set to 128k tokens. Therefore, depending on the number of characters of the prompt input by the questioner, a situation has occurred where it conflicts with the limit and the answer required by the questioner cannot be obtained. As a technology for dealing with such a situation, for example, in Patent Document 1, a technology for creating a response content by switching a bot according to the content of a message in a message application is disclosed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In order to avoid the above-described situation, the generative AI acquires hint information for generating an answer to a prompt by using vectorization and a vector database, and generates an answer by setting this hint information and the prompt together. However, this vector database also has limitations. For example, one vector database has a maximum file size of 256 MB (megabytes), and the total maximum size is set to around tens of gigabytes (gigabytes). Therefore, it is difficult to use generative AI on big data. On the other hand, generative AI cannot obtain appropriate hint information without performing similarity searches using the vector database, which leads to a decrease in the accuracy of the answers. As a result, when vectorizing big data, users of generative AI were forced to either perform the vectorization process and vector database creation process from scratch themselves, preparing an optimal vector database, or compromise on the accuracy of the answers. Therefore, when responding to big data, there is a need for technology that can improve response accuracy without requiring the preparation of an optimal vector database. However, even with the technology described in Patent Document 1, since it only involves switching bots, it was not possible to improve the accuracy of responses if there was a risk of situations that violated the restrictions. Therefore, the inventor focused on a mechanism that appropriately combines multiple bots to improve the accuracy of responses to prompts from questioners.

[0005] In view of these challenges, the present invention aims to provide a prompt engineering computer, a prompt engineering method, and a program that can improve the accuracy of responses when providing responses to big data without the need to prepare an optimal vector database. [Means for solving the problem]

[0006] The present invention is a prompt engineering computer that uses generative AI to output a response to a prompt, A prompt acquisition unit that acquires the aforementioned prompt, A decision unit that determines two or more bots to respond to the aforementioned prompt, A question data generation unit that generates question data for the bot, Based on the aforementioned question data, a question unit that asks questions to each of the bots, The response acquisition unit acquires each of the responses generated by each of the aforementioned bots, A response document generation unit generates a response document that combines each of the above responses as a response to the prompt, An output unit that outputs the aforementioned response document as a response to the prompt, We provide a prompt engineering computer equipped with the necessary features.

[0007] According to the present invention, by linking two or more bots and using a response document compiled from the responses obtained from each bot as a response to big data, it becomes possible to improve the accuracy of responses without the need to prepare an optimal vector database.

[0008] Although this invention falls under the category of computers, it will produce similar functions and effects in other categories such as methods and programs. [Effects of the Invention]

[0009] According to the present invention, it is possible to improve the accuracy of responses. [Brief explanation of the drawing]

[0010] [Figure 1] This is a diagram illustrating the overview of the prompt engineering system 1. [Figure 2] This diagram shows the functional configuration of the prompt engineering system 1. [Figure 3] This diagram shows a flowchart of the question processing performed by the prompt engineering computer 10. [Figure 4] This diagram shows a flowchart of the substitution process performed by the prompt engineering computer 10. [Figure 5] This diagram shows a flowchart of the redecision process performed by the prompt engineering computer 10. [Figure 6] It is a diagram showing the flowchart of the end process executed by the prompt engineering computer 10. [Figure 7] It is a diagram showing the flowchart of the answer process executed by the prompt engineering computer 10. [Figure 8] It is a diagram schematically showing the breakdown of each bot when the prompt engineering computer 10 acquires a prompt related to student life.

Embodiments for Carrying Out the Invention

[0011] Hereinafter, embodiments for carrying out the present invention (hereinafter referred to as embodiments) will be described in detail with reference to the accompanying drawings. In the following figures, the same elements are denoted by the same numbers or symbols throughout the description of the embodiments.

[0012] [Overview of Prompt Engineering System 1] FIG. 1 is a schematic diagram for explaining the overview of the prompt engineering system 1. Based on FIG. 1, the components of the prompt engineering system 1 will be described. The prompt engineering system 1 is a system comprising at least a prompt engineering computer 10 having a server function and outputting an answer to a prompt using generative AI. In the present embodiment, the prompt engineering system 1 includes, in addition to the prompt engineering computer 10, a questioner terminal 3 used by a questioner.

[0013] The questioner terminal 3 is, for example, a terminal device such as a mobile phone, a smartphone, a tablet terminal, a personal computer, a laptop computer, etc. The number of the questioner terminals 3 may be any number corresponding to the number of questioners, and is not particularly limited, and can be designed as appropriate. The prompt engineering computer 10 has server functions and may be implemented, for example, by a single computer or by multiple computers such as a cloud computer. The cloud computer in this specification may be either one that uses any computer in a scalable manner when performing a specific function or one that includes a plurality of functional modules to realize a certain system and freely combines and uses its functions. In addition to the above-described questioner terminal 3 and prompt engineering computer 10, the prompt engineering system 1 may include other terminals, devices, etc., and the number, type, and functions thereof are not particularly limited and can be appropriately designed.

[0014] An overview of the processing steps when the prompt engineering system 1 outputs an answer to a prompt using generative AI will be described.

[0015] The prompt engineering computer 10 acquires a prompt (step S1). The prompt engineering computer 10 acquires the prompt input by the questioner from the questioner terminal 3.

[0016] The prompt engineering computer 10 determines two or more bots that will answer the prompt (step S2). The prompt engineering computer 10 can execute a plurality of bots. The plurality of bots include a first bot that exchanges messages with the questioner terminal 3, a second bot that has learned the respective learning contents of the plurality of bots excluding the first bot, and two or more bots (third bot, fourth bot,..., nth bot) that have learned different learning contents and answer based on the question data from the second bot. The two or more bots that answer each have an index indicating their own learning content, and the second bot holds this index. The prompt engineering computer 10 inputs the prompt obtained by the first bot into the second bot, the second bot refers to an index and obtains two or more bot candidates that will provide the identified answer, and from these bot candidates, it determines two or more bots that will respond to the prompt.

[0017] The prompt engineering computer 10 generates question data for the bot (step S3). The prompt engineering computer 10 generates question data based on the question sentences included in the prompt, in accordance with the learning content of each determined bot.

[0018] The prompt engineering computer 10 asks each bot a question based on the question data (step S4). The prompt engineering computer 10 inputs the question sentences from the generated question data into the corresponding bots and asks each bot a question.

[0019] The prompt engineering computer 10 retrieves each response generated by each bot (step S5). The prompt engineering computer 10 retrieves the answers that each bot has generated based on the question entered into it.

[0020] The prompt engineering computer 10 generates a response document that combines each response as an answer to the prompt (step S6). The prompt engineering computer 10 combines the responses obtained from each bot to generate a response document that answers the prompt.

[0021] The prompt engineering computer 10 outputs the response document as the answer to the prompt (step S7). The prompt engineering computer 10 outputs the generated response document to the questioner terminal 3, causing the questioner terminal 3 to display this response document.

[0022] The above is an overview of the prompt engineering system 1. According to this prompt engineering system 1, when responding to big data, it is possible to improve the accuracy of responses without having to prepare an optimal vector database.

[0023] [Device configuration] Figure 2 is a block diagram showing the configuration of the prompt engineering system 1. Based on Figure 2, the device configuration of the prompt engineering system 1 will be described. The prompt engineering system 1 consists of at least a prompt engineering computer 10 that outputs a response to a prompt using a generation AI. In this embodiment, the prompt engineering system 1 consists of the prompt engineering computer 10 and a questioner terminal 3. The prompt engineering system 1 is a system in which the prompt engineering computer 10 is connected to the questioner terminal 3 via a network 8 such as the Internet, a company LAN (Local Area Network), Wi-Fi, or a VPN (Virtual Private Network), enabling data communication. In the prompt engineering system 1, the number of questioner terminals 3 can be designed as appropriate according to the number of questioners, and is not particularly limited. Furthermore, the prompt engineering system 1 may include other terminals and devices in addition to the questioner terminals 3 and the prompt engineering computer 10, and the number, type, and function of these other terminals and devices can be designed as appropriate.

[0024] Questioner's device 3 is a terminal device used by the questioner, and can be a mobile phone, smartphone, tablet, personal computer, laptop computer, etc. The questioner's terminal 3 includes a CPU (Central Processing Unit), GPU (Graphics Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), etc., as a terminal control unit, and a communication unit that includes devices to enable communication with other terminals and devices. The questioner terminal 3 is equipped with various devices as an input / output unit that receive predetermined inputs, input and output various types of data, etc.

[0025] The prompt engineering computer 10 has server functionality and may be implemented on a single computer, for example, or on multiple computers, like a cloud computer. The prompt engineering computer 10 is an information processing device that creates prompts to be input into a large-scale language model. The prompt engineering computer 10 includes a CPU, GPU, RAM, ROM, etc. as a control unit, and a communication unit that includes a device for communicating with other terminals and devices, a prompt acquisition unit for acquiring prompts, and an output unit for outputting response documents as responses to prompts. The prompt engineering computer 10 includes a data storage unit, such as a hard disk, semiconductor memory, recording media, or memory card, as its storage unit. The prompt engineering computer 10 includes, as a processing unit, various devices for executing various processes, a decision unit for determining two or more bots to respond to a prompt, a question data generation unit for generating question data for the bots, a question unit for asking each bot a question based on the question data, an answer acquisition unit for acquiring each of the answers generated by each bot, and an answer document generation unit for generating an answer document that compiles each answer as a response to the prompt.

[0026] In the prompt engineering computer 10, the control unit loads a predetermined program and, in cooperation with the communication unit, realizes a prompt acquisition module and an output module. Furthermore, in the prompt engineering computer 10, the control unit loads a predetermined program, and in cooperation with the memory unit, realizes a memory module. Furthermore, in the prompt engineering computer 10, the control unit loads a predetermined program and, in cooperation with the processing unit, realizes a decision module, a question data generation module, a question module, an answer acquisition module, an acquisition determination module, a replacement module, a count determination module, a termination module, and an answer document generation module.

[0027] The following describes each process performed by the prompt engineering system 1, along with the processes performed by the modules mentioned above. In this specification, each module may perform its processing as a function of its own, or it may perform it through a predetermined application.

[0028] [Question processing performed by the prompt engineering computer 10] The question processing performed by the prompt engineering computer 10 will be explained based on Figure 3. This figure is a flowchart of the question processing performed by the prompt engineering computer 10. The question processing consists of the following steps: prompt acquisition processing to obtain a prompt (step S1), decision processing to determine two or more bots to provide an answer to the prompt (step S2), question data generation processing to generate question data for the bots (step S3), question processing to ask each bot a question based on the question data (step S4), and answer acquisition processing to obtain each answer generated by each bot (step S5).

[0029] The prompt acquisition module acquires the prompt (step S10). A prompt is a question, explanation, instruction, or summary for the questioner to use the generated AI. It must at least include a question, but it does not necessarily have to include explanations, instructions, or summaries. Furthermore, a prompt relates to a predetermined topic, which may be, for example, a predetermined item (daily necessities, electronic devices, etc.), a patient's treatment method (treatment for chronic back pain, etc.), or student life (courses, curriculum, career support, etc.), and is not particularly limited. The prompt acquisition module obtains the prompt from the questioner's terminal 3. The prompt acquisition module obtains prompts from the questioner's terminal 3 via a bot (hereinafter also referred to as the first bot) that communicates with the questioner through dialogue. The first bot is one that has been pre-stored by the memory module. The questioner terminal 3 receives prompt input from the questioner through a predetermined UI (User Interface) for inputting prompts, in a dialogue format with the first bot. The questioner terminal 3 accepts input such as text input and voice input, and receives prompt input from the questioner in a dialogue format such as exchanging messages with the first bot. The questioner terminal 3 sends the received prompt to the prompt engineering computer 10. The prompt acquisition module receives this prompt and retrieves it. The prompt acquisition module will then retrieve the prompt that the first bot has received input from.

[0030] The decision module determines two or more bots to respond to the prompt (step S11). The decision module, via the first bot, presents the question content included in the acquired prompt to a bot (hereinafter also referred to as the second bot) that determines two or more bots that will provide answers, which have been previously memorized by the memory module, and instructs the second bot to respond with a candidate bot capable of answering this question. The second bot is a bot that has previously learned what each of two or more bots (excluding the first and second bots) that the memory module has memorized has learned, and is a bot that possesses an index that shows the characteristics (learned content) of each of the two or more bots (hereinafter simply referred to as multiple bots) excluding the first and second bots. The content that each of the multiple bots has learned and the learning method can be designed as appropriate and are not particularly limited. The total number of multiple bots can also be designed as appropriate and are not particularly limited. In this embodiment, the memory module is described as storing n bots, namely the third bot, the fourth bot, the fifth bot, ..., the nth bot. Each of the multiple bots is a bot that has learned different content. Each of the multiple bots can also be said to be a bot that has learned a corresponding category, meaning that the third bot has learned the first category, the fourth bot has learned the second category, the fifth bot has learned the third category, ..., the nth bot has learned the mth category. The decision module inputs the question content included in the prompt to the second bot via the first bot. The second bot analyzes this question content, refers to an index indicating the characteristics of each bot, and, via the first bot, responds with the bot candidate that it deems most relevant. In this embodiment, the second bot is described as responding with the third bot, the fourth bot, and the fifth bot as bot candidates. The decision module determines, based on the responses, two or more bots that will actually answer the question from the questioner. The decision module selects three bots, the third bot, the fourth bot, and the fifth bot, as the bots that will provide the answer. The decision module only needs to select at least two bots to be the bots that will provide answers. In other words, the decision module only needs to select at least a third bot that has learned the first category and a fourth bot that has learned the second category as the bots that will provide answers.

[0031] The question data generation module generates question data for the bot (step S12). Question data may be generated based on the question content included in the acquired prompt, or some or all of the question content included in the acquired prompt may be replaced with the question content corresponding to each determined bot, and the data based on the replaced question content may be generated as question data. If the question content is replaced, question data for each bot should be generated, such as question data for the third bot, question data for the fourth bot, and question data for the fifth bot. Furthermore, the question data generated by the question data generation module is not limited to the examples described above. It may generate question data common to all bots, or it may generate question data corresponding to some bots (for example, question data corresponding to the third and fourth bots) and question data corresponding to the other bots.

[0032] The question module asks each bot a question based on the question data (step S13). The question module presents the generated question data to each of the two or more selected bots via the first bot and instructs them to answer the question data. The question module then presents the generated question data to each of the third, fourth, and fifth bots via the first bot and instructs them to answer the question data. The question module inputs question data to each of the bots—the 3rd bot, the 4th bot, and the 5th bot—via the 1st bot. By inputting the question data to each bot, the question module then asks a question to each of the bots.

[0033] The response acquisition module retrieves each response generated by each bot (step S14). The answer acquisition module retrieves the answers generated by each bot from each bot that has been input with question data. The third bot, fourth bot, and fifth bot each generate an answer to the input question data based on their own learned content. The third bot, fourth bot, and fifth bot each output the generated answer to the answer acquisition module via the first bot. The responses generated by each bot and retrieved by the response acquisition module include not only the answer itself, but also hint information for generating the answer. The response acquisition module receives the responses generated and output by the third, fourth, and fifth bots via the first bot, and acquires each of the responses generated by each bot.

[0034] The above is the question processing procedure. If the prompt engineering computer 10 fails to obtain an answer from a bot during question processing, it performs the replacement process described later.

[0035] [Replacement process executed by prompt engineering computer 10] The substitution process performed by the prompt engineering computer 10 will be explained based on Figure 4. This figure is a flowchart showing the substitution process performed by the prompt engineering computer 10.

[0036] The acquisition determination module determines whether or not it was possible to obtain a response from at least one bot (step S20). The acquisition determination module determines whether or not it was unable to obtain an answer from at least one of the third bot, fourth bot, and fifth bot during the processing of step S14 described above. The acquisition determination module may determine whether or not it was unable to obtain an answer based on whether or not it was unable to obtain the answer itself from each bot, or based on whether or not it was unable to obtain an answer from each bot within a predetermined time, or based on whether or not it found that the answer obtained from each bot contained a predetermined response sentence (such as an answer sentence containing phrases indicating that an answer was impossible to provide, such as "unknown," "I don't know," or "cannot answer"). The acquisition determination module determines that it was unable to obtain an answer if it was unable to obtain the answer itself, if it was unable to obtain an answer within a predetermined time, or if the obtained answer contained a predetermined response sentence, and determines that it was able to obtain an answer otherwise. If the acquisition determination module has been able to obtain responses from all bots (step S20 NO), the prompt engineering computer 10 terminates this replacement process and executes the response processing described later. The prompt engineering computer 10 uses the acquired responses to execute the response processing described later.

[0037] On the other hand, if the acquisition determination module fails to acquire a response from at least one bot (step S20 YES), the replacement module generates question data by replacing the question sentence in the question data with a different question sentence (step S21). The replacement module replaces the question text in the question data for bots that failed to provide an answer with a different question text, generating the replaced question data. The replacement module determines which question text prevented the bot from generating an answer and identifies the wording in the question text that the bot cannot answer. The replacement module then replaces the identified wording with wording that the bot can answer.

[0038] The question module then asks questions to bots that were unable to provide answers based on the replaced question data (step S22). The question module, via the first bot, presents the replaced question data to bots that were unable to obtain an answer, and instructs them to answer this replaced question data. In this embodiment, the case where an answer could not be obtained from the third bot is described as an example. The question module, via the first bot, presents the replaced question data to the third bot and instructs it to answer this replaced question data. The question module inputs the replaced question data into the third bot via the first bot. By inputting the question data into the third bot, the question module will ask the bots that were unable to provide an answer.

[0039] The response retrieval module retrieves responses generated by bots that were unable to provide a response (step S23). The answer retrieval module retrieves the answers generated by the bot from the bot that was input with the substituted question data. The third bot generates answers to the input substituted question data based on its own learning. The third bot outputs the generated answers to the answer retrieval module via the first bot. The response acquisition module receives responses generated and output by the third bot via the first bot, and acquires responses generated by bots that were unable to provide responses.

[0040] The above is the substitution process. Although the above-described replacement process assumes that a response could not be obtained from bot 1, the present invention is also applicable even if responses could not be obtained from multiple bots. In such cases, the prompt engineering computer 10 can simply execute the processing from step S21 onwards for each of the multiple bots. If the prompt engineering computer 10 fails to obtain an answer from any bots during the replacement process, it will perform a re-determination process, as described later.

[0041] [Redecision process performed by prompt engineering computer 10] The re-determination process performed by the prompt engineering computer 10 will be described based on Figure 5. This figure is a flowchart of the re-determination process performed by the prompt engineering computer 10.

[0042] The acquisition determination module determines whether or not it was possible to obtain an answer from the bot that asked the question, based on the replaced question data (step S30). The acquisition determination module determines whether or not it was unable to obtain an answer from the third bot during the processing in step S23 described above. The acquisition determination module may determine whether or not it was unable to obtain an answer based on whether or not it was unable to obtain the answer itself from the bot that asked the question based on the replaced question data, or based on whether or not it was unable to obtain an answer from the bot that asked the question based on the replaced question data within a predetermined time, or based on whether or not the answer obtained from the bot that asked the question based on the replaced question data contains a predetermined answer sentence (such as an answer sentence containing phrases indicating that an answer was impossible to provide, such as "unknown," "I don't know," or "cannot answer"). The acquisition determination module determines that it was unable to obtain an answer if it was unable to obtain the answer itself, if it was unable to obtain an answer within a predetermined time, or if the obtained answer contains a predetermined answer sentence, and determines that it was able to obtain an answer otherwise. If the acquisition determination module is able to obtain an answer from the bot that asked the question based on the replaced question data (step S30 NO), the prompt engineering computer 10 terminates this re-determination process and executes the answer processing described later. The prompt engineering computer 10 uses the acquired answer to execute the answer processing described later.

[0043] On the other hand, if the acquisition determination module is unable to obtain an answer from the bot that asked the question based on the replaced question data (step S30 YES), the decision module determines another bot that will provide the answer (step S31). The decision module, via the first bot, presents the question content of the replaced question data to the second bot and instructs it to provide another bot candidate that can answer this question. At this time, the decision module instructs the second bot to provide another bot candidate that can answer the question from among the remaining bots, excluding the bots that have already provided an answer. Since the decision module has already received answers from the fourth and fifth bots, it instructs the second bot to provide another bot candidate that can answer the question from among the remaining bots, excluding the third bot (which could not provide an answer) and the fourth and fifth bots (which have already provided answers). The decision module inputs the question content of the replacement question data to the second bot via the first bot. The second bot analyzes this question content and, after excluding the third bot (which could not obtain an answer via the first bot) and the fourth and fifth bots (which had already obtained answers), it provides a candidate bot that can answer the question from among the remaining bots. In this embodiment, the sixth bot is described as the bot candidate that provides the answer. The decision module determines, based on the response, which bot will answer the question in the replaced question data. The decision module then determines that the sixth bot will be another bot that will provide the answer. Furthermore, if the decision module receives a response from the first bot stating that no other bot candidates exist, it terminates the re-decision process. At this time, if there are two or more bots from which responses have already been obtained, the prompt engineering computer 10 uses the obtained responses to perform the response processing described later. On the other hand, if there are not two or more bots from which responses have already been obtained, the prompt engineering computer 10 generates a response document stating that a response is not possible when generating the response document during the response processing described later. Furthermore, if the decision module receives a response from the first bot indicating that there are multiple other bot candidates, it may select one bot from among the candidates, or it may select multiple bots. In this case, the decision module may select a number of bot candidates that is within a predetermined number, or it may select bot candidates that meet certain conditions (such as bots with similar learning content to bots that did not respond).

[0044] The question data generation module generates question data for the bot (step S32). The question data generation module generates question data by replacing the question data that was asked to the bot from which no answer was obtained, and using that as question data for the other bot that was determined. The question data generation module may use the question data generated by the process in step S21 as question data for the other bot, or it may perform the same process as in step S21 to generate new question data for the other bot.

[0045] The question module asks the bot a question based on the question data (step S33). The question module presents the generated question data to another bot, selected via the first bot, and instructs it to answer the question data. The question module then presents the question data to the sixth bot, also via the first bot, and instructs it to answer the question data. The question module inputs question data to the sixth bot via the first bot. By inputting question data to the sixth bot, the question module is essentially asking a question to another bot.

[0046] The response retrieval module retrieves the response generated by the bot (step S34). The answer retrieval module retrieves answers generated by another bot from another bot that has been input with question data. The sixth bot generates answers to the input question data based on its own learning. The sixth bot outputs the generated answers to the answer retrieval module via the first bot. The response acquisition module receives responses generated and output by the 6th bot via the 1st bot, and acquires responses generated by other bots.

[0047] The above is the re-determination process. Although the above-described re-determination process assumes that a response could not be obtained from bot 1, the present invention is also applicable even if responses could not be obtained from multiple bots. In such cases, the prompt engineering computer 10 can simply execute the processes from step S31 onward for each of the multiple bots. If the prompt engineering computer 10 fails to obtain a response from any bots during the re-determination process, it will perform the termination process described later.

[0048] [Termination process executed by prompt engineering computer 10] The termination process performed by the prompt engineering computer 10 will be described based on Figure 6. This figure is a flowchart showing the termination process performed by the prompt engineering computer 10.

[0049] The acquisition determination module determines whether or not it was able to obtain an answer from another bot (step S440). The acquisition determination module determines whether or not it was unable to obtain an answer from the sixth bot during the processing in step S34 described above. The acquisition determination module may determine whether or not it was unable to obtain an answer based on whether or not it was unable to obtain the answer itself from another bot that asked a question based on the question data, or based on whether or not it was unable to obtain an answer from another bot that asked a question based on the question data within a predetermined time, or based on whether or not it was unable to obtain an answer from another bot that asked a question based on the question data. If the acquisition determination module is able to obtain an answer from another bot that asked a question based on the question data (step S40 NO), the prompt engineering computer 10 terminates this termination process and uses the acquired answer to perform the answer processing described later. Furthermore, if the acquisition determination module has determined multiple different bots during the re-determination process, it executes the process in step S40 for each of those bots. If the prompt engineering computer 10 fails to obtain a response from at least one bot, it executes the process in step S41, described later, for the bots from which it failed to obtain a response.

[0050] On the other hand, if the acquisition determination module fails to obtain an answer from another bot that asked a question based on the question data (step S40 YES), the count determination module determines whether the number of times an answer could not be obtained is greater than or equal to a predetermined number (step S41). The count determination module determines whether the number of times an answer could not be obtained as a result of the re-determination process described above is equal to or greater than a predetermined number. If the count determination module determines that the number is less than the predetermined number (step S41 NO), the prompt engineering computer 10 terminates the termination process and executes the processes from step S31 onwards in the re-determination process described above again. Furthermore, if the count determination module has determined multiple different bots during the re-determination process, it executes the process in step S41 for each of those bots. If the prompt engineering computer 10 has failed to obtain a response from at least one bot for a predetermined number of times or more, it executes the process in step S42, described later, for the bots from which it failed to obtain a response.

[0051] On the other hand, if the count determination module determines that the number of times is greater than or equal to a predetermined number (step S41 YES), the termination module terminates the generation of question data (step S42). The termination module ends the generation of question data so as not to perform the replacement and re-determination processes described above. As a result, if the prompt engineering computer 10 fails to obtain a response from the first bot more than a predetermined number of times, it stops presenting question data to another bot and instructing it to answer this question data. In other words, if the prompt engineering computer 10 fails to obtain a response from a bot more than a predetermined number of times, it terminates the process itself without determining another bot again. Furthermore, if the termination module has determined multiple different bots during the re-determination process, it will execute the process in step S42 for each of those bots.

[0052] The above completes the termination process. The prompt engineering computer 10 can perform substitution and re-determination processes from at least two bots until it finally obtains an answer. However, if it fails to obtain an answer from at least two bots even after performing substitution and re-determination processes a predetermined number of times or more, it will generate an answer document indicating that it is impossible to provide an answer during the answer processing described later. Furthermore, if the prompt engineering computer 10 has obtained responses from at least two bots by the time the termination process is completed, it will use the obtained responses to perform the response processing described later.

[0053] [Answer processing performed by prompt engineering computers 10] Based on Figure 7, the response processing performed by the prompt engineering computer 10 will be described. This figure is a flowchart of the response processing performed by the prompt engineering computer 10. The response processing consists of a response document generation process (step S6) which generates a response document that compiles each response as a response to a prompt, and an output process (step S7) which outputs the response document as a response to a prompt.

[0054] The response document generation module generates a response document that combines each response as an answer to the prompt (step S50). The response document generation module uses each response obtained from the bot as hint information to present the question content from the questioner included in the obtained prompt to the generating AI, and instructs it to answer this question. The generating AI generates an answer to the question from the questioner based on hint information and the question content, and outputs the generated answer to the answer document generation module. The response document generation module generates a response document based on this response and creates a compiled response document that addresses the prompt.

[0055] The output module outputs the response document as the answer to the prompt (step S51). The output module sends the response document to the questioner terminal 3 as an answer to the prompt via the first bot. The questioner terminal 3 receives and displays the response document sent by the prompt engineering computer 10 via a predetermined UI. The output module outputs the response document as an answer to the prompt by displaying the response document on the questioner's terminal 3 via a predetermined UI.

[0056] The above is the response process.

[0057] Specific examples of each of the processes described above will be explained. First, as a first specific example, we will explain the processing content when the prompt engineering computer 10 receives a prompt related to student life as predetermined content, referring to Figure 8. Each bot stored in the prompt computer 10 has learned information related to categories of student life such as university, faculty, and career support. Furthermore, the questioner is a student. The diagram schematically shows the breakdown of each bot category when the prompt engineering computer 10 obtains prompts related to student life.

[0058] In step S10, the prompt acquisition module obtains the prompt "Tell me about the data science course" from the questioner terminal 3 via the first bot (not shown) as a prompt related to student life.

[0059] In step S11, the decision module, in order to determine two or more bots to respond to this prompt, presents the question "Tell me about data science courses" to the second bot 20 and instructs it to select bot candidates that can answer this question. The second bot analyzes the question, refers to an index of multiple bots, and selects bot candidates that have the characteristics to answer the question. The decision module selects two or more bots to provide answers based on the bot candidates provided. In the diagram, the decision module selects the third bot 21, the fourth bot 22, and the fifth bot 23 to provide answers. The third bot 21 is a bot that has learned information about the Computer Science department category, the fourth bot 22 is a bot that has learned information about the Statistics department category, and the fifth bot 23 is a bot that has learned information about the Career Support category.

[0060] In step S12, the question data generation module generates question data for the bots. For example, the question data generation module generates a question for the third bot 21 that the third bot 21 can answer, such as "Tell me about the subject," based on the prompt. The question data generation module also generates a question for the fourth bot 22 that the fourth bot 22 can answer, such as "Tell me about the statistics and data analysis courses," based on the prompt. The question data generation module also generates a question for the fifth bot 23 that the fifth bot 23 can answer, such as "Tell me about careers after graduation," based on the prompt.

[0061] In step S13, the question module asks the corresponding bots the generated question data. The question module asks the third bot 21, "Tell me about the subject," the fourth bot 22, "Tell me about the statistics and data analysis courses," and the fifth bot 23, "Tell me about careers after graduation."

[0062] In step S14, the response acquisition module acquires the responses generated by the third bot 21, the fourth bot 22, and the fifth bot 23, respectively.

[0063] Here, the prompt engineering computer 10 performs a substitution process, and if it is unable to obtain a response from at least one of the third bot 21, fourth bot 22, and fifth bot 23, it replaces the question text in the question data for the bot that could not provide a response with a different question text, and then asks the corresponding bot the substituted question data. For example, if the question module is unable to obtain a response from the third bot 21, it replaces the question text in the question data for the third bot 21, "Tell me about the department," with another question text that the third bot 21 can answer to the original prompt (e.g., "Tell me about the basic courses and curriculum of the Data Science department"). The question module asks the third bot 21 a question based on the replaced question data, and retrieves the answer that the third bot 21 generates for the question text of this replaced question data.

[0064] Furthermore, the prompt engineering computer 10 performs a re-determination process, and if it is unable to obtain a response from the third bot 21, the decision module determines another bot that will provide the response. If the decision module determines the sixth bot (not shown) as the other bot, it asks the sixth bot the question text of the replaced question data that was asked to the third bot 21 in the replacement process, the question text of the question data before replacement, or the question text of the question data generated to suit the sixth bot, and obtains the response that the sixth bot generates in response to the question text of this question data.

[0065] Furthermore, the prompt engineering computer 10 performs termination processing and, if it fails to obtain a response from the sixth bot and fails to obtain a predetermined number of responses, it terminates the generation of question data.

[0066] In step S50, the response document generation module uses the responses obtained from the third bot 21, the fourth bot 22, and the fifth bot 23 as hint information to present the question content from the questioner included in the obtained prompt to the generating AI, and instructs it to answer this question. The response document generation module generates response documents based on the responses generated by the generation AI.

[0067] In step S51, the output module outputs the generated response document to the questioner terminal 3.

[0068] The above are specific examples of each process in the first example.

[0069] Next, a second specific example will be given illustrating the processing steps taken when the prompt engineering computer 10 obtains a prompt regarding a predetermined item. In this second example, the case where the item is a daily necessities (such as adhesive) will be described. Each bot stored in the prompt computer 10 has learned information about the categories of goods and services offered, such as the various tenants and products handled within the shopping mall. The questioner is a customer of the shopping mall. Detailed explanations of the same content as the first specific example described above will be omitted.

[0070] In step S10, the prompt acquisition module obtains a prompt from the questioner terminal 3 regarding a specified item (e.g., "adhesive for bonding shoe soles").

[0071] In step S11, the decision module, in order to determine two or more bots that will respond to this prompt, presents the question "adhesive for bonding shoe soles" to the second bot and instructs it to select bot candidates that can answer this question. The second bot analyzes the question, refers to the index of multiple bots, and selects bot candidates that have the characteristics to answer the question. The decision module selects two or more bots to provide answers based on the bot candidates provided. The decision module then selects the third and fourth bots to provide answers. The third bot is a bot that has learned information about the categories in the daily necessities section, and the fourth bot is a bot that has learned information about the shoe repair category.

[0072] In steps S12 to S14, the prompt engineering computer 10 generates question data for each bot, asks the corresponding bot the question data, and retrieves the answers generated by each bot. Here as well, the prompt engineering computer 10 performs replacement processing, re-determination processing, and termination processing if necessary.

[0073] In steps S50 and S51, the prompt engineering computer 10 uses the answers obtained from each bot as hint information to present the question content from the questioner included in the obtained prompt to the generation AI, instructs it to answer this question, generates an answer document generated by the generation AI, and outputs the generated answer document to the questioner terminal 3.

[0074] The above are specific examples of each process in the second example.

[0075] Next, a third example will be described of the processing that occurs when the prompt engineering computer 10 obtains a prompt for a specific item different from the second example, as predetermined content. In the third example, the case where the item is a computer component (CPU, etc.) will be described. Each bot stored by the prompt computer 10 has learned information about the electronic device itself, such as the components that make up the electronic device (computer, etc.), the part numbers, and the category of the electronic device, such as the vulnerabilities of the electronic device. The questioner is the user of the electronic device. Detailed explanations of the same content as the examples mentioned above will be omitted.

[0076] In step S10, the prompt acquisition module obtains a prompt from the questioner terminal 3 regarding a specified item (e.g., "Regarding vulnerabilities in the 7th generation CPU").

[0077] In step S11, the decision module, in order to determine two or more bots that will respond to this prompt, presents the question "Regarding vulnerabilities in the 7th generation CPU" to the second bot and instructs it to respond with candidate bots capable of answering this question. The second bot analyzes the question, refers to an index of multiple bots, and responds with candidate bots that have the characteristics to answer the question. The decision module selects two or more bots to provide answers based on the bot candidates provided. The decision module then selects the third and fourth bots to provide answers. The third bot is one that has learned information about motherboard categories, and the fourth bot is one that has learned information about device driver categories.

[0078] In steps S12 to S14, the prompt engineering computer 10 generates question data for each bot, asks the corresponding bot the question data, and retrieves the answers generated by each bot. Here as well, the prompt engineering computer 10 performs replacement processing, re-determination processing, and termination processing if necessary.

[0079] In steps S50 and S51, the prompt engineering computer 10 uses the answers obtained from each bot as hint information to present the question content from the questioner included in the obtained prompt to the generation AI, instructs it to answer this question, generates an answer document generated by the generation AI, and outputs the generated answer document to the questioner terminal 3.

[0080] The above are specific examples of each process in the third example.

[0081] Next, as a fourth specific example, we will explain the processing content when the prompt engineering computer 10 obtains a prompt regarding a patient's treatment method as predetermined content. Each bot stored in the prompt computer 10 has learned information about the categories of patient treatment methods, such as each medical department and pharmacy. The questioner is the patient receiving treatment. Detailed explanations of the same content as the examples mentioned above will be omitted.

[0082] In step S10, the prompt acquisition module obtains a prompt from the questioner terminal 3 regarding the patient's own treatment method (e.g., "treatment method for chronic lower back pain").

[0083] In step S11, the decision module, in order to determine two or more bots to respond to this prompt, presents the question "Treatment for chronic lower back pain" to the second bot and instructs it to select bot candidates that can answer this question. The second bot analyzes the question, refers to the index of multiple bots, and selects bot candidates that have the characteristics to answer the question. The decision module selects two or more bots to provide responses based on the bot candidates provided. The decision module then selects the third, fourth, and fifth bots to provide responses. The third bot is one that has learned information related to the orthopedic category, the fourth bot is one that has learned information related to the rehabilitation category, and the fifth bot is one that has learned information related to the pharmacy category.

[0084] In steps S12 to S14, the prompt engineering computer 10 generates question data for each bot, asks the corresponding bot the question data, and retrieves the answers generated by each bot. Here as well, the prompt engineering computer 10 performs replacement processing, re-determination processing, and termination processing if necessary.

[0085] In steps S50 and S51, the prompt engineering computer 10 uses the answers obtained from each bot as hint information to present the question content from the questioner included in the obtained prompt to the generation AI, instructs it to answer this question, generates an answer document generated by the generation AI, and outputs the generated answer document to the questioner terminal 3.

[0086] The above are specific examples of each process in the fourth example.

[0087] Although the processes described above are listed as separate processes, the prompt engineering computer 10 can also be configured to execute some or all of the above processes in combination. Furthermore, the prompt engineering computer 10 can be configured to execute each process at times other than those described.

[0088] The means and functions described above are realized by a computer (including the CPU, information processing unit, and various terminals) reading and executing a predetermined program. The program may be provided, for example, via a network from the computer (SaaS: Software as a Service) or as a cloud service. Alternatively, the program may be provided in a form recorded on a computer-readable recording medium. In this case, the computer reads the program from the recording medium, transfers it to an internal or external recording device, records it, and executes it. Alternatively, the program may be pre-recorded on a recording device (recording medium) and provided to the computer from that recording device via a communication line.

[0089] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above. Furthermore, the effects described in the embodiments of the present invention are merely a list of the most preferred effects arising from the present invention, and the effects of the present invention are not limited to those described in the embodiments of the present invention.

[0090] A first aspect disclosed in this embodiment is a prompt engineering computer that uses a generation AI to output a response to a prompt, A prompt acquisition unit that acquires the aforementioned prompt, A decision unit that determines two or more bots to respond to the aforementioned prompt, A question data generation unit that generates question data for the bot, Based on the aforementioned question data, a question unit that asks questions to each of the bots, The response acquisition unit acquires each of the responses generated by each of the aforementioned bots, A response document generation unit generates a response document that combines each of the above responses as a response to the prompt, An output unit that outputs the aforementioned response document as a response to the prompt, We provide a prompt engineering computer equipped with the necessary features.

[0091] A second aspect disclosed in this embodiment is that the determination unit determines the bot by referring to an index indicating the characteristics of the bot. The present invention provides a prompt engineering computer as described in the first embodiment.

[0092] A third aspect disclosed in this embodiment is that if the question data generation unit is unable to obtain a response from at least one bot, it generates question data in which the question sentence of the question data is replaced with a different question sentence. The aforementioned questioning unit asks questions to bots that were unable to obtain answers based on the replaced question data. The present invention provides a prompt engineering computer as described in the first embodiment.

[0093] A fourth aspect disclosed in this embodiment is that if the decision unit is unable to obtain an answer from the bot that asked the question about the replaced question data, it determines another bot that will provide the answer. A prompt engineering computer according to a third embodiment is provided.

[0094] A fifth aspect disclosed in this embodiment is that the prompt acquisition unit acquires a prompt relating to a predetermined content, The determination unit determines two or more bots, including at least one bot for the first category and one for the second category, in response to a prompt regarding the predetermined content. The present invention provides a prompt engineering computer as described in the first embodiment.

[0095] A sixth aspect disclosed in this embodiment is that the predetermined content is a predetermined article, The first category mentioned above is the place where the goods are sold, The second category is a location related to the article. A prompt engineering computer according to a fifth aspect is provided.

[0096] A seventh aspect disclosed in this embodiment is that the predetermined content is a predetermined article, The first category is a component of the article, The second category is the vulnerability of the component. A prompt engineering computer according to a fifth aspect is provided.

[0097] An eighth aspect disclosed in this embodiment is a treatment method for a patient, The previous category 1 is the medical department. The second category mentioned above is what you can find at a pharmacy. A prompt engineering computer according to a fifth aspect is provided.

[0098] The ninth aspect disclosed in this embodiment is that the predetermined content is student life, The first category mentioned above is the academic discipline. The second category mentioned above is career support. A prompt engineering computer according to a fifth aspect is provided. [Explanation of symbols]

[0099] 1. Prompt Engineering System 3. Questioner's terminal 8 Networks 10 Prompt Engineering Computer 20 2nd Bot 21 3rd Bot 22 4th Bot 23 Bot No. 5

Claims

1. A prompt engineering computer that uses generative AI to output answers to prompts, A prompt acquisition unit that acquires the aforementioned prompt, A decision unit that determines two or more bots to respond to the aforementioned prompt, A question data generation unit that generates question data for the bot, Based on the aforementioned question data, a question unit that asks questions to each of the bots, The response acquisition unit acquires each of the responses generated by the aforementioned bots, A response document generation unit generates a response document that combines each of the above responses as a response to the prompt, An output unit that outputs the aforementioned response document as a response to the prompt, A prompt engineering computer equipped with [features / equipment].

2. The determination unit refers to an index indicating the characteristics of the bot and determines the bot. The prompt engineering computer according to claim 1.

3. If the question data generation unit fails to obtain a response from at least one bot, it generates question data in which the question sentence of the question data is replaced with a different question sentence. The aforementioned questioning unit asks questions to bots that were unable to obtain answers based on the replaced question data. The prompt engineering computer according to claim 1.

4. The determination unit, if it is unable to obtain an answer from the bot that asked the question using the replaced question data, determines another bot to provide the answer. The prompt engineering computer according to claim 3.

5. The prompt acquisition unit acquires a prompt relating to a predetermined content, The determination unit determines two or more bots, including at least one bot related to the first category and one related to the second category, in response to a prompt relating to the predetermined content. The prompt engineering computer according to claim 1.

6. The specified contents are specified articles, The first category mentioned above is the place where the goods are sold, The second category is a location related to the article. The prompt engineering computer according to claim 5.

7. The specified contents are specified articles, The first category is a component of the article, The second category is the vulnerability of the component. The prompt engineering computer according to claim 5.

8. The aforementioned specified content is the treatment method for the patient, The first category mentioned above is the medical department. The second category mentioned above is what you can find at a pharmacy. The prompt engineering computer according to claim 5.

9. The aforementioned specified content is student life, The first category mentioned above is the subject, The second category mentioned above is career support. The prompt engineering computer according to claim 5.

10. A prompt engineering method performed by a prompt engineering computer that outputs a response to a prompt using generative AI, The steps include obtaining the aforementioned prompt, The steps include determining two or more bots that will respond to the aforementioned prompt, The steps include generating question data for the bot, Based on the aforementioned question data, the step of asking each of the aforementioned bots a question, The steps include obtaining each of the answers generated by each of the aforementioned bots, The steps include generating a response document that combines each of the above responses as a response to the above prompt, The steps include outputting the aforementioned response document as a response to the prompt, A prompt engineering method that includes the following features.

11. Using generative AI, a prompt engineering computer outputs answers to prompts. Steps to obtain the aforementioned prompt, The step of determining two or more bots that will respond to the aforementioned prompt, A step of generating question data for the bot, Based on the aforementioned question data, the step of asking each of the aforementioned bots a question, Steps to obtain each of the answers generated by each of the aforementioned bots, A step of generating a response document that combines each of the above responses as a response to the above prompt, A step of outputting the aforementioned response document as a response to the prompt, A computer-readable program for executing a command.

Images