Prompt generation device, report generation device, system, and program

Abstract

This prompt generation device which generates a prompt to a language model comprises: an acquisition unit which acquires information summarizing a malfunction state; and a generation unit which generates a prompt including said information and an instruction text for the coordination of a report creation role, an information collection role, and a report checking role.

Description

Prompt generation device, report generation device, system, and program

[0001] The present invention relates to the technical field of network maintenance and operation.

[0002] When congestion or equipment failure occurs during network operation, it can cause problems such as temporary difficulty for users in using services or being unable to connect to the network. When this happens, network operators must prepare a network failure report (hereinafter referred to as a "failure report") summarizing the nature of the failure and its cause, and post it on a webpage or other means to inform general users and related companies of the situation. In the case of a serious accident with a large impact, a report must also be submitted to the Ministry of Internal Affairs and Communications, and additional reports may need to be prepared.

[0003] There is a common format for failure reports, and each item must be filled in. Specifically, the items include the date and time of the failure, the main affected service, the affected area, the number of people affected, the cause, and the recovery status (Non-Patent Document 1, Non-Patent Document 2).

[0004] The task of creating a fault report involves gathering the information that needs to be recorded and writing that information in an easy-to-understand format. Previous report examples are useful when creating a fault report. Currently, the collection of information that needs to be recorded and the creation of report content are left to humans. For example, Non-Patent Document 3 summarizes fault response methods, but does not mention automating or streamlining report creation. When it comes to information collection, network administrators may use their own appropriate tools.

[0005] Ministry of Internal Affairs and Communications report, "Status of Accidents in Telecommunications Services (FY2022)," https: / / www.soumu.go.jp / main_content / 000897675.pdf Ministry of Internal Affairs and Communications website: "Outline of the Accident Reporting System," https: / / www.soumu.go.jp / menu_seisaku / ictseisaku / net_anzen / jiko / index.html "Lessons Learned from High-Reliability Information Processing Systems," March 16, 2020, https: / / www.ipa.go.jp / publish / qv6pgp0000000wo6-att / 000071987.pdf

[0006] Although failure reports need to be developed and submitted quickly, it is difficult to gather the information required for a failure report quickly. Furthermore, if the reporter is not familiar with creating a failure report, they must refer to past reports, which incurs personnel costs. Furthermore, the information must be updated periodically until the failure is resolved, which incurs additional costs in determining the timing of updates and corrections.

[0007] The present invention has been made in view of the above points, and an object of the present invention is to provide a technique for automatically generating a failure report.

[0008] According to the disclosed technology, there is provided a prompt generation device that generates a prompt for a language model, the prompt generation device including: an acquisition unit that acquires information summarizing the status of a disability; and a generation unit that generates the prompt including the information and an instruction statement for coordinating a report creator, an information collector, and a report checker.

[0009] The disclosed technology makes it possible to automatically generate a failure report.

[0010] 1 is a configuration diagram of a failure report generation device 100. FIG. 2 is a diagram for explaining the operation of the failure report generation device 100. FIG. 3 is a flowchart showing a failure report generation procedure. FIG. 4 is a diagram showing an image of processing within an LLM. FIG. 5 is a diagram showing a sequence of failure report generation. FIG. 6 is a configuration diagram of a prompt generation device 200. FIG. 7 is a configuration diagram of a report generation device 300. FIG. 8 is a diagram showing an example of the hardware configuration of the device.

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

[0012] (Overview of Technology Relating to Embodiment) The technology relating to the present embodiment relates to the maintenance and operation of various IT systems. When a network failure occurs, a failure report generation device 100 (described later) acquires information about the failure, such as the cause of the failure, the location of the failure, and the extent of the damage, and performs processing to generate a failure report based on this information.

[0013] Specifically, the problem report generation device 100 uses a large-scale language model (e.g., ChatGPT) that has made remarkable achievements as a sentence generation model, inputs known information, collects information on unknown report items, and automatically generates the required report format content from information on past reports. The key points of this technology are as follows.

[0014] It is difficult to obtain a report corresponding to an actual network fault case from a large-scale language model. A basic approach to this problem is to provide the large-scale language model with information specific to the actual network fault case in the "prompt" that serves as input data to the large-scale language model. However, it is difficult to determine what information should be provided so that the large-scale language model can create an appropriate report for a specific fault case. In this embodiment, the fault report generation device 100 provides appropriate information to the large-scale language model.

[0015] In addition, by assigning roles such as report writer, information gatherer, and report reviewer to large-scale language models and linking them together, the risk of generating false information, which is a concern when tasks are performed without any specifications, is reduced.

[0016] In this embodiment, a large language model (LLM) is used as the language model, but a language model other than what is called a "large language model" may also be used. The "large language model" is an example of a language model.

[0017] The configuration and operation of the device / system according to this embodiment will be described in detail below.

[0018] 1 shows an example of the configuration of a failure report generation device 100 according to this embodiment. When a network failure exceeding a certain threshold is observed in a monitored network, the failure report generation device 100 executes processing to create a failure report that describes the failure content, cause, recovery status, etc. To this end, the failure report generation device 100 includes an information aggregation unit 110, a related report extraction unit 120, and a prompt generation unit 130.

[0019] FIG. 2 is a diagram showing the functions of the failure report generation device 100, as well as the functions that are the source of information input to the failure report generation device 100 and the functions that are the destination of information output from the failure report generation device 100.

[0020] Specifically, the diagram shows a user usage monitoring tool 10, a fault information monitoring tool 20, a past report DB (database) 30, an LLM 40, and an administrator 50. The "administrator 50" is more specifically a terminal or the like operated by an administrator (person). The LLM 40 is more specifically a device (computer) equipped with the LLM. The LLM 40 may be provided within the fault report generation device 100.

[0021] A prompt is input to the LLM 40 from the prompt generation unit 130. The prompt is data in text format. The text is a sequence of multiple sentences. The content of the prompt has a significant impact on the quality of the output result from the LLM 40, so the prompt must include information necessary for creating a fault report.

[0022] In this embodiment, the information aggregating unit 110, the related report extracting unit 120, and the prompt generating unit 130 generate information to be included in the prompt.

[0023] (Operation of Each Unit) An overview of the operation of each unit in the problem report generation device 100 will be described with reference to FIG.

[0024] <Information aggregation unit 110> The information aggregation unit 110 receives as input the fault status data obtained from the fault information monitoring tool 20, and generates and outputs "A. Fault status aggregation text" which expresses the status of the data in text data format so that the prompt can recognize it.

[0025] In addition, the information aggregation unit 110 receives as input the "data in which user usage status is stored in chronological order" obtained from the user usage status monitoring tool 10, and generates and outputs "B. Usage status aggregation text" which is expressed in text data format so that the status of the data can be recognized by the prompt.

[0026] <Related report extraction unit 120> The related report extraction unit 120 receives as input "A. Fault status summary text" output by the information aggregation unit 110 and past reports stored in the past report DB 30, and outputs the top k past reports with the highest relevance to the currently occurring fault as "C. Related report list."

[0027] By having the LLM 40 refer to past reports that are highly relevant, it is expected that a more appropriate report will be generated.

[0028] <Prompt generation unit 130> As shown in Figure 2, the prompt generation unit 130 receives as input data six types of data: the above information A to information C, "D. Network configuration and setting information," "E. Action history information," and "F. Report approval requirements."

[0029] "D. Network configuration and setting information" is data that describes in text format the configuration diagram of the target network, the bandwidth setting values ​​of each link, routing settings, etc.

[0030] "E. Action history information" is data on the actions taken in response to the failure that occurred, either automatically output from the monitoring system or manually converted into text.

[0031] "F. Report Approval Requirements" is data that describes in text format the rules that must be observed when creating a report.

[0032] When generating a prompt, the prompt generator 130 defines three roles - a report creator, an information collector, and a report checker - in addition to the six types of data described above, and includes in the prompt a statement instructing cooperation between the three roles in order to improve the accuracy of the output results from the LLM 40. The instruction statements defining the three roles are also provided as prompts.

[0033] (Processing Flow) The processing procedure executed by the problem report generation device 100 and the LLM 40 will be described below along the procedure of the flowchart in Fig. 3. Here, input data and output data will be described in more detail.

[0034] <S101 (Step 101)> When information is obtained in the fault information monitoring tool 20 (e.g., internal system information such as a QoS monitoring tool and system logs, or external system information such as SNS) that suggests a network fault has occurred, such as a predetermined quality measurement item exceeding a threshold or complaints from multiple users being recognized, the fault report generation device 100 is started. Note that "starting" may also be rephrased as "starting processing."

[0035] The information aggregation unit 110 acquires information from the fault information monitoring tool 20 and stores the acquired information in chronological order. The time length of the time series data is set to any time length between a certain time before the fault occurs and a certain time after the fault occurs.

[0036] Similarly, the information aggregation unit 110 observes internal information such as packet fluctuations and fluctuations in the number of users using the user usage monitoring tool 10, and saves the information before and after the failure report generation device 100 is started in chronological order.

[0037] <S102> The information aggregating unit 110 uses the information input from the fault information monitoring tool 20 and saved to generate and output "A. Trouble situation summary text", which is text explaining the fault situation.

[0038] The information aggregator 110 generates "A. Fault Status Summary Text" from information obtained from the fault information monitoring tool 20, for example, using a common text template. For example, templates such as "The delay on link XX changed from XX msec to XX msec" and "An alarm XXX was output on device XX" are provided, and the information aggregator 110 generates "A. Fault Status Summary Text" by substituting specific values ​​for the XX and XXX. Specific examples of "A. Fault Status Summary Text" are as follows. Note that "A. Fault Status Summary Text" may also be obtained from the fault information monitoring tool 20. ---------------------------------------------------------------- Information A: March 13, 15:00 Link A went down. 15:00 -> 15:10 Delay changed from 10 ms to 100 ms. 15:00 -> 15:10 Throughput dropped from 1 Gbps to 20 Mbps. --------------------------------------------------- Similarly, the information aggregation unit 110 uses the packet fluctuation status and user usage count fluctuation status obtained and saved from the user usage status monitoring tool 10 to generate and output "B. Usage status aggregation text", which is text that explains the packet fluctuation status and user usage count fluctuation status.

[0039] For example, a template such as "The number of users in area A changed from ● to △" is provided, and the information aggregator 110 generates "B. Usage status summary text" by substituting specific values ​​for ● and △. Specific examples of "B. Usage status summary text" are as follows. Note that "B. Usage status summary text" may also be obtained from the user usage monitoring tool 10. ------------------------------------------------ Information B: March 13, 14:50 -> 15:00 The number of users in area A increased from 1,000 to 4,000. The number of sessions increased from 20,000 to 80,000. ------------------------------------------------ <S103> In S103, the related report extractor 120 generates and outputs "C. Related report list" based on "A. Fault status summary text" and the information read from the past report DB 30. Specifically, it is as follows:

[0040] First, we will explain the information stored in the past report DB 30. Each past report in the past report DB 30 is stored as a pair of two texts: a failure report text (p_i) that describes the actual failure, and information summary text (q_i) that explains the network status at the time. Specific examples of failure report text (p_i) and information summary text (q_i) are shown below. ---Information summary text (q_i)--- Information A: March 13, 15:00 Link A went down. 15:00 -> 15:10 Latency changed from 10 msec to 100 msec. 15:00 -> 15:10 Throughput dropped from 1 Gbps to 20 Mbps. ---Information summary text (q_i) ends here--------------- ---Failure report text (p_i)--------------- Network Failure Report 1. Date and time of the failure: March 13, 2024, 2:50 PM 2. Scope of the failure: - Internet connectivity was intermittently lost throughout the company. - The mail server and file sharing service became unavailable. 3. Symptoms of the failure: - Numerous users reported being unable to connect. - A significant decrease in network speed was observed. 4. Cause: - Initial investigations indicated a malfunction in the router firmware. - The system attempted to automatically restart due to an overload condition, but was unsuccessful. 5. Solution and Response: - The network team urgently restarted the router and updated its firmware. - We are planning to install additional network equipment to increase redundancy. 6. Recovery time: - All services were confirmed to have been restored approximately two hours after the failure occurred. 7. Measures to prevent recurrence: - We will strengthen regular monitoring and maintenance of routers and other network equipment. - Consider additional measures to improve network redundancy and fault tolerance. ---Failure report text (p_i) ends here---If each report is represented as r_i = (p_i, q_i), the information stored in the past report DB 30 can be expressed as {r1, r2, ...rn}.

[0041] The related report extraction unit 120 vectorizes q_i using a text distributed representation tool (such as Illamax index) for {r1, r2, ... rn} read from the past report DB 30. The vectorized results are represented as vr1, vr2, ... vrn. Similarly, a distributed representation is generated for "A. Fault status summary text" using the above tool and vectorized. The vectorized result is represented as u. These are vectors of the same length as vri.

[0042] The related report extraction unit 120 calculates the cosine similarity α as follows, and selects k reports with the largest value (similar reports).

[0043] The associated report extraction unit 120 outputs the selected k failure report texts (p_i) as “C. Associated report list.” Specific examples of failure report texts (network failure reports) are as shown above.

[0044] <S104> In S104, the prompt generation unit 130 receives as input "A. Fault status summary text," "B. Usage status summary text," "C. Related report list," "D. Network configuration and setting information," "E. Action history information," and "F. Report approval requirements," and creates a prompt that will be input data to the LLM 40. As the prompt, the prompt generation unit 130 first writes the above information A to F in order, and also writes an instruction statement (coordination instruction) for coordinating the three roles of the report creator, information collector, and report checker.

[0045] <S105> In S105, the prompt generation unit 130 inputs the prompt generated in S104 to the LLM 40. The LLM 40 generates and outputs a problem report in accordance with the prompt. An example of the problem report is a problem report stored in the past report DB 30 described above. The problem report output from the LLM 40 is sent to, for example, the administrator 50. The problem report output from the LLM 40 may be input to the problem report generation device 100, and the problem report generation device 100 may notify the administrator 50 of the problem report.

[0046] An example of a prompt including an instruction statement generated by the prompt generation unit 130 is shown below. ---Prompt Example--- <Information on the Failure Status and User Usage> Information A: March 13th, 15:00 Link A went down. 15:00 -> 15:10 Latency changed from 10 msec to 100 msec. 15:00 -> 15:10 Throughput decreased from 1 Gbps to 20 Mbps. Information B: March 13th, 14:50 -> 15:00 The number of users in Area A increased from 1,000 to 4,000. The number of sessions increased from 20,000 to 80,000. Information C: Network Failure Report 1. Date and Time of Failure: 14:45, March 13, 2024 2. Situation before the failure: - The system was operating normally, and no particular abnormalities were reported. - Daily backup processing was being performed. 3. Scope of the failure: ............ . Information D: Information E: ............ . ............ . == ... [Information Collector:] - The report creator will ask questions about the problem situation, so please answer them. When answering, please refer to A, B, D, and E. - If you are unable to answer the questions at this time, please reply that you are currently investigating. [Report Checker:] - Check the format of the report created by the report creator. Please refer to F when checking. If it is insufficient, please instruct the report creator on what to correct.- Once there are no problems with the report submitted by the report creator, please give them "approval" and upload the report to the web page. -----------------End of example prompt-----------------An example prompt is as shown above.

[0047] Figure 4 shows an image of the processing within the LLM 40 after the above prompt is given. As shown in Figure 4, three parties create a problem report through conversation according to the three roles defined in the prompt. The conversation between the three parties is included in the response content of the LLM 40. Note that the report may be output by the report creator, as in the example shown in Figure 4, or by the report checker.

[0048] (Processing Sequence) The processing sequence described above will now be described with reference to the sequence chart shown in Fig. 5. Fig. 5 also shows step numbers (S101 to S105) corresponding to the flowchart shown in Fig. 3.

[0049] In S11, the information aggregating unit 110 acquires fault information from the fault information monitoring tool 20. In S12, the information aggregating unit 110 acquires user usage status from the user usage status monitoring tool.

[0050] In S13, the administrator 50 sends a problem report generation instruction to the prompt generation unit 130. The problem report generation instruction includes information D, E, and F shown in FIG.

[0051] In S14, the prompt generation unit 130 requests a situation summary text from the information collection unit 110. In S15, the information collection unit 110 transmits "A. Failure situation summary text" to the prompt generation unit 130. In S16, the information collection unit 110 transmits "B. Usage situation summary text" to the prompt generation unit 130.

[0052] In S17, the prompt generation unit 130 requests a related report from the related report extraction unit 120. In S18, the related report extraction unit 120 requests a failure status summary text from the information aggregation unit 110. In S19, the information aggregation unit 110 transmits "A. Failure status summary text" to the related report extraction unit 120.

[0053] The related report extraction unit 120 generates "C. Related report list" using "A. Trouble situation summary text" and past reports read from the past report DB 30, and sends "C. Related report list" to the prompt generation unit 130 in S20.

[0054] In S21, the prompt generation unit 130 generates a prompt. Examples of the prompt are as described above. In S22, the prompt generation unit 130 sends the generated prompt to the LLM 40. In S23, the LLM 40 generates a problem report. In S24, the LLM 40 sends the problem report to the administrator 50.

[0055] (Other Device Configuration Examples) The device configuration is not limited to the configuration shown in FIG. 1. For example, the information aggregating unit 110, the related report extracting unit 120, and the prompt generating unit 130 may each be a single device (computer). The information aggregating unit 110, the related report extracting unit 120, and the prompt generating unit 130 may be called an information aggregating device, a related report extracting device, and a prompt generating device, respectively. Furthermore, the prompt generating unit 130 (prompt generating device) may include the functions of either or both of the information aggregating unit 110 and the related report extracting unit 120. Furthermore, the LLM 40 may be called a report generating device.

[0056] Furthermore, any of the information aggregating device, related report extracting device, prompt generating device, and report generating device may be referred to as an information processing device. Furthermore, a system may be configured that includes a prompt generating device and a report generating device.

[0057] 6 shows an example of the configuration of the prompt generation device 200. As shown in FIG. 6, the prompt generation device 200 includes an acquisition unit 210 and a generation unit 220.

[0058] The acquiring unit 210 acquires information that summarizes the failure status. The acquiring unit 210 may acquire the information A, information B, information C, information D, information E, and information F described above.

[0059] The generator 220 generates and outputs a prompt including the information and instructions for coordinating the report creator, information collector, and report checker.

[0060] Fig. 7 shows an example of the configuration of the report generation device 300. As shown in Fig. 7, the report generation device 300 has an input unit 310, an output unit 320, and a language model 330. The language model 330 is, for example, an LLM. The word model 330 is, for example, a neural network.

[0061] The input unit 310 receives the prompt generated by the prompt generation device 200 and inputs the prompt to the language model 330. The output unit 320 receives the report generated by the language model 330 and outputs the report.

[0062] The language model 330 receives as input a prompt containing information summarizing the status of the failure and instructions for coordinating the report creator, information gatherer, and report checker, and generates a report about the failure by having the report creator, information gatherer, and report checker converse with each other in accordance with the prompt.

[0063] (Hardware Configuration Example) Any of the devices described in this embodiment (fault report generation device, information aggregation device, related report extraction device, prompt generation device, report generation device, information processing device, etc.) can be realized, for example, by causing a computer to execute a program. This computer may be a physical computer or a virtual machine on the cloud.

[0064] That is, the device can be realized by executing a program corresponding to the processing performed by the device using hardware resources such as a CPU and memory built into a computer. The program can be recorded on a computer-readable recording medium (such as a portable memory) and stored or distributed. The program can also be provided via a network such as the Internet or email.

[0065] Fig. 8 is a diagram showing an example of the hardware configuration of the computer. The computer in Fig. 8 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, all of which are interconnected via a bus B. The computer may further include a GPU.

[0066] The program that realizes the processing on the computer is provided by a recording medium 1001, such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, but may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files, data, etc.

[0067] The memory device 1003 reads and stores a program from the auxiliary storage device 1002 when an instruction to start the program is received. The CPU 1004 realizes functions related to the device in accordance with the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) or the like according to the program. The input device 1007 is composed of a keyboard, mouse, buttons, a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the results of calculations.

[0068] (Summary, Effects, etc. of the Embodiment) As described above, the technology described in this embodiment makes it possible to automatically create a failure report based on the actual failure details. Information missing at the time of creation can be automatically supplemented by having the LLM respond. Furthermore, when using the LLM, by defining and assigning roles to a report creator, information collector, and report checker, and establishing a system in which these roles cooperate and check each other, it is expected that the risk of generating false information will be reduced compared to when no roles are specified.

[0069] In other words, the technology described in this embodiment provides benefits such as "there is no need to manually collect information to fill in the items in the report" and "because it is generated automatically, it can accommodate a variety of formats, which leads to reduced human costs and reduced work time required for report creation."

[0070] The following additional notes are provided regarding the above-described embodiments.

[0071] <Additional Notes> (Additional Item 1) A prompt generation device that generates a prompt for a language model, comprising: an acquisition unit that acquires information summarizing the status of a failure; and a generation unit that generates the prompt including the information and an instruction statement for coordinating a report creator, an information gatherer, and a report checker. (Additional Item 2) The prompt generation device of Additional Item 1, wherein the prompt instructs the report creator to create a report about the failure, the information gatherer to answer questions from the report creator, and the report checker to instruct report corrections and report approval. (Additional Item 3) The prompt generation device of Additional Item 1, wherein the prompt includes a past report related to the failure. (Supplementary Item 4) A report generation device comprising a language model, wherein the language model receives as input a prompt including information summarizing the status of a failure and an instruction for coordinating a report creator, an information gatherer, and a report checker, and generates a report about the failure by having a conversation between the report creator, the information gatherer, and the report checker in accordance with the prompt. (Supplementary Item 5) A system including a prompt generation device that generates a prompt for a language model and a report generation device comprising the language model, wherein the prompt generation device comprises: an acquisition unit that acquires information summarizing the status of a failure, and a generation unit that generates the prompt including the information and an instruction for coordinating a report creator, an information gatherer, and a report checker, wherein the language model in the report generation device receives as input the prompt, and generates a report about the failure by having a conversation between the report creator, the information gatherer, and the report checker in accordance with the prompt.(Supplementary Item 6) A non-transitory storage medium storing a program for causing a computer to function as a prompt generation device that generates prompts for a language model, the program causing the computer to function as: an acquisition unit that acquires information summarizing the status of a failure, and a generation unit that generates the prompt including the information and an instruction for coordinating a report writer, an information gatherer, and a report checker. (Supplementary Item 7) A non-transitory storage medium storing a program for causing a computer to function as a language model in a report generation device, the language model receiving as input a prompt including information summarizing the status of a failure and an instruction for coordinating a report writer, an information gatherer, and a report checker, and generating a report about the failure by causing a conversation between the report writer, the information gatherer, and the report checker in accordance with the prompt.

[0072] Although the present embodiment has been described above, the present invention is not limited to such a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0073] 10 User usage status monitoring tool 20 Failure information monitoring tool 30 Past report DB 40 LLM 50 Administrator 100 Failure report generation device 110 Information aggregation unit 120 Related report extraction unit 130 Prompt generation unit 200 Prompt generation device 210 Acquisition unit 220 Generation unit 2 300 Report generation device 310 Input unit 320 Output unit 330 Language model 1000 Drive device 1001 Recording medium 1002 Auxiliary storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims

1. A prompt generation device that generates prompts for a language model, comprising: an acquisition unit that acquires information summarizing the status of a disability; and a generation unit that generates the prompt including the information and an instruction for coordinating a report creator, an information collector, and a report checker.

2. The prompt generation device of claim 1, wherein the prompt instructs the report creator to create a report about the problem, the information gatherer to answer questions from the report creator, and the report checker to instruct report corrections and approval.

3. The prompt generating device of claim 1, wherein the prompt includes a past report related to the problem.

4. A report generation device having a language model, wherein the language model receives as input prompts including information summarizing the status of a failure and instructions for coordinating a report creator, an information gatherer, and a report checker, and generates a report about the failure by having the report creator, the information gatherer, and the report checker converse with each other in accordance with the prompts.

5. A system including a prompt generation device that generates a prompt for a language model, and a report generation device equipped with the language model, wherein the prompt generation device includes: an acquisition unit that acquires information summarizing the status of a failure; and a generation unit that generates the prompt including the information and an instruction for coordinating a report creator, an information gatherer, and a report checker, and the language model in the report generation device receives the prompt as an input, and generates a report about the failure by having a conversation between the report creator, the information gatherer, and the report checker in accordance with the prompt.

6. A program for causing a computer to function as a prompt generation device that generates prompts for a language model, the program causing the computer to function as: an acquisition unit that acquires information summarizing the status of a failure; and a generation unit that generates the prompt including the information and instructions for coordinating a report creator, an information collector, and a report checker.

7. A program for causing a computer to function as a language model in a report generation device, wherein the language model receives as input prompts including information summarizing the status of a failure and instructions for coordinating a Report Writer, Information Gatherer, and Report Checker, and generates a report about the failure by having the Report Writer, Information Gatherer, and Report Checker converse with each other in accordance with the prompts.

Images