Information processing device, information processing method, and information processing program

Abstract

[Problem] A prompt input to a trained AI model includes a plurality of processes but is created as a single block of text. Accordingly, when an inappropriate response is returned, there arises a problem that it is difficult to determine at which processing stage the cause of the inappropriate response lies. [Solution] In order to solve the above problem, there are provided an information processing device, etc. comprising: a prompt setting unit that sets a prompt; and a divided prompt generation unit that inputs the set prompt to a trained AI model to generate divided prompts obtained by dividing the set prompt.

Description

Information processing device, information processing method, and information processing program

[0001] The present invention relates to an information processing device that acquires desired information from a file data group containing various information using a trained AI model.

[0002] In recent years, the use of large-scale language models (hereinafter sometimes abbreviated as LLM) to acquire predetermined information from a group of file data has become widespread. For example, Patent Literature 1 discloses a technology that uses LLM to collect character information related to predetermined items from a variety of printed materials with no fixed format, such as advertising flyers.

[0003] Patent No. 7430437

[0004] The technology of Patent Document 1 collects one or more pieces of character information relating to one or more specified items from a group of character strings printed on a flyer read by OCR by providing the LLM with prompts including instructions for specifying the items to be collected and various other instructions related to the collection.

[0005] As in the example above, in most cases, a prompt given to an LLM contains instructions for multiple processes leading up to the final desired information. For example, a prompt such as "Please create a table showing the average life expectancy for each prefecture in the Kanto region" contains multiple processes, such as "Extract prefectures in the Kanto region," "Obtain the average life expectancy for each prefecture," and "Create a table of the results."

[0006] However, even though multiple processes are included, the prompts entered are often created as a single sentence as described above. Therefore, if the results obtained from the created prompt are undesirable or incorrect information is returned, it is difficult to determine which stage of the process is causing the problem and which process needs to be corrected to improve the situation.

[0007] Therefore, in order to solve the above-mentioned problems, the present invention provides the following information processing device etc. That is, the information processing device includes a prompt setting unit that sets a prompt, and a split prompt generation unit that inputs the set prompt to a trained AI model and generates split prompts by dividing the set prompt.

[0008] In addition to the above features, the present invention provides an information processing device further comprising a divided prompt setting unit that sets the generated divided prompt.

[0009] In addition to the above features, the present invention provides an information processing device that further has a division processing step generation unit that inputs the generated division prompt or the set division prompt into the trained AI model and generates a division processing step corresponding to the generated division prompt or the set division prompt.

[0010] In addition to the above features, the present invention provides an information processing device further comprising a division processing step result acquisition unit that acquires the processing results of the generated division processing steps.

[0011] In addition to the above features, the split prompt generation unit provides an information processing device that inputs the set prompt into the trained AI model and generates split prompts by splitting the set prompt, together with the corresponding split processing steps.

[0012] In addition to the above features, the present invention provides an information processing device that further has a sample setting unit that sets correct answer sample file data that is input to the trained AI model together with the set prompt and serves as a correct answer example for obtaining the answer required by the set prompt, and correct answer information contained in the correct answer sample file data.

[0013] In addition to the above features, the present invention provides an information processing device further comprising a correct answer information explanation prompt setting unit that sets a correct answer information explanation prompt that explains the correct answer information that has been set.

[0014] In addition to the above features, the split prompt generation unit provides an information processing device that inputs the set correct answer sample file data and correct answer information together with the set prompt into a trained AI model to generate a split prompt.

[0015] In addition to the above features, the splitting processing step generation unit provides an information processing device that inputs the set correct sample file data and correct answer information into the trained AI model to generate splitting processing steps.

[0016] In addition to the above features, the information processing device further includes a split processing step evaluation unit that compares the processing result of the split processing step generated by inputting the set correct answer sample file data and correct answer information into the trained AI model, or the split processing step generated from a split prompt generated by inputting the set correct answer sample file data and correct answer information into the trained AI model, with the correct answer information, and evaluates the generated split processing step.

[0017] The present invention also provides an information processing device having a prompt setting unit that sets a prompt, and a processing step generation unit that inputs the set prompt into the trained AI model and generates divided processing steps based on the set prompt.

[0018] In addition to the above features, the present invention provides an information processing device further comprising a processing step result acquisition unit that acquires the processing results of the generated processing steps.

[0019] In addition to the above features, the present invention provides an information processing device that further has a processing step prompt generation unit that inputs the generated processing step into a trained AI model and generates a processing step prompt that instructs the trained AI model to use the generated processing step.

[0020] The present invention also provides an information processing method executed by an information processing device, which includes a prompt setting step for setting a prompt, and a split prompt generation step for inputting the set prompt into a trained AI model and generating split prompts by dividing the set prompt.

[0021] Also provided is an information processing program that causes an information processing device to execute a prompt setting step for setting a prompt, and a split prompt generation step for inputting the set prompt into a trained AI model and generating split prompts by dividing the set prompt.

[0022] The present invention also provides an information processing method executed by an information processing device, which includes a prompt setting step for setting a prompt, and a processing step generation step for inputting the set prompt into the trained AI model and generating divided processing steps based on the set prompt.

[0023] Also provided is an information processing program that causes an information processing device to execute a prompt setting step of setting a prompt, and a processing step generation step of inputting the set prompt into the trained AI model and generating divided processing steps based on the set prompt.

[0024] This invention makes it possible to clarify the prompts input into a trained AI model and the intermediate processing that leads to the answer obtained from that input, allowing for review and correction to obtain the desired answer.

[0025] A block diagram showing an example of the functional configuration of an information processing apparatus according to a first embodiment. A conceptual diagram showing an example of a division prompt. A conceptual diagram showing an example of a viewer that displays correct sample file data. A conceptual diagram showing an example of the configuration of hardware that realizes the information processing apparatus according to the first embodiment. A flow diagram simply showing an example of the processing flow of the information processing apparatus according to the first embodiment. A block diagram showing an example of the functional configuration of an information processing apparatus according to a second embodiment. A flow diagram simply showing an example of the processing flow of the information processing apparatus according to the second embodiment.

[0026] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the present invention should not be limited to these embodiments and can be embodied in various forms without departing from the spirit and scope of the present invention.

[0027] <Example 1> <Overview> The present invention is an information processing device that inputs set prompts into a trained AI model and generates prompts for each intermediate processing step included in the prompts. This makes it possible to clarify the intermediate processing leading up to the final result, and to judge the validity of the prompts for each step and the accuracy of the intermediate results, as well as to review and correct each step to obtain a more accurate answer.

[0028] The following describes the functions and processing flow of the information processing device, as well as the hardware. The functional blocks of the system described below can be implemented as a combination of hardware and software. Specifically, if a computer is used, these include hardware components such as a central processing unit (CPU), main memory, bus, or secondary storage device (such as a hard disk drive, nonvolatile memory, or storage media such as CDs or DVDs, and their reader drives), input devices used for information input, printers, display devices, and other external peripheral devices, as well as interfaces for those external peripheral devices, communication interfaces, driver programs and other application programs for controlling the hardware, and user interface applications. The CPU processes data input from input devices and other interfaces and stored in memory or on a hard disk in accordance with a program loaded in main memory, processing and storing the data, and generating commands to control the hardware and software. Alternatively, the functional blocks of the system can be implemented using dedicated hardware.

[0029] Furthermore, this invention can be realized not only as a system but also as a method. Furthermore, a part of such an invention can be configured as software. Furthermore, a program used to cause a computer to execute such software, and a recording medium on which the program is fixed, are naturally included within the technical scope of this invention (the same applies throughout this specification).

[0030] <Functional Configuration> Fig. 1 is a block diagram showing an example of the functional configuration of an information processing apparatus according to Example 1. As shown in Fig. 1, the information processing apparatus 100 includes a prompt setting unit 101, a divided prompt generating unit 102, a divided prompt setting unit 103, a divided processing step generating unit 104, and a divided processing step result acquiring unit 105.

[0031] In this embodiment, the trained AI model is a machine learning model that has been trained in advance to handle general-purpose tasks. A typical example is the well-known LLM, but it may also be a different model such as a small language model or a multimodal language model that can handle images and other formats.

[0032] <Prompt Setting Unit> The prompt setting unit 101 has a function of setting a prompt. A prompt is an instruction to a trained AI model, and is used to extract and acquire information such as numerical values, character strings, image data, and audio data from target file data, and to calculate and process such information to obtain, as an answer from the trained AI model, information (total number, average information, maximum information, coordinate information, spatial relationships such as distance and collision, 3D model information, 3D graphic information, image, specific part of an image, image content, character information, etc.).

[0033] The target file data may be stored in a storage device provided in the information processing device, or may be stored in a storage device connected to the information processing device via a communication line, etc. The file data may be in a variety of file formats, including, but not limited to, Word files, Portable Document Format (PDF) files, Excel files, image files, Comma Separated Values ​​(CSV) files, text files, Building Information Modeling (BIM) files, 2D CAD (Computer Aided Design) files, and 3D model files.

[0034] For example, when targeting CSV files related to the facilities of multiple buildings and obtaining the "total number of elevators" from the file data, the desired information would be the number of elevators in each building extracted from each file and the sum of those numbers. In addition to numerical calculations, the desired information could also be the results of processing such as three-dimensional spatial processing or two-dimensional graphic processing. For example, when targeting three-dimensional model files of multiple buildings and obtaining the "collision points between columns and beams," the desired information would be the three-dimensional spatial information obtained by extracting all column and beam models included in each file, calculating whether or not there are any collisions, and extracting the collision points.

[0035] Furthermore, for example, if you are looking for "inconsistencies between elevations and floor plans" in two-dimensional drawing files of multiple buildings, you can extract all of the architectural design information for elevations and floor plans contained in each file, check whether there are any inconsistencies when they are reconstructed into three-dimensional solids, and then document the inconsistencies into text, which becomes the desired information.

[0036] The prompt to be set is expressed in natural language, for example, "Tell me the distance between pillars." Such a prompt may be input by the user, or may be set in part or in whole by the information processing device, or may be set again by the user after modifying what has been set by the information processing device.

[0037] Prompts can not only instruct the trained AI model on the desired answer, but can also include conditions for guiding the model to an appropriate answer and examples to refer to. For example, when setting a prompt to instruct the model to extract a "column object" that is part of a BIM model, the prompt can be set to a sentence such as, "Please extract the value of 'column object'. 'Column object' is part of a building's BIM model and has a long, three-dimensional structure that extends vertically from the ground."

[0038] Furthermore, the prompt is not limited to one, and multiple prompts can be set. For example, in addition to the instruction to extract "column objects" mentioned above, instructions to extract "beam objects" and "elevator type and number" can also be set. For "beam object" extraction, the prompt can be set using a sentence such as, "Please extract the value of 'beam object'. 'Beam object' is part of the building's BIM model, and has a three-dimensional structure that is horizontally long and located more than 2 meters from the ground." For "elevator type and number," the prompt can be set using a sentence such as, "Please extract the elevator type and number in the BIM model. If there are no applicable items, leave blank."

[0039] <Split prompt generation unit> The split prompt generation unit 102 has a function of inputting the set prompt to a trained AI model and generating split prompts by dividing the set prompt. The set prompt may be an instruction that leads to an answer through multiple processes, and the split prompt generation unit generates the split prompt by using the trained AI model to infer instructions for each of the multiple processes included in the prompt.

[0040] Figure 2 is a conceptual diagram showing an example of a segmented prompt. Here, the prompt "Tell me the distance between pillars" 201 is shown. When this prompt is input to the trained AI model, the trained AI model infers the processes included in the input prompt and generates segmented prompts corresponding to each process.

[0041] For example, as shown in the figure, if the set prompt is "Find the distance between the columns" 201, the step-by-step processes to be performed to find the distance between the columns are inferred, and according to each inferred process, a split prompt "Find the columns" 202, a split prompt "Find the surface of each column" 203, and a split prompt "Draw a line connecting the surfaces and find its length" 204 are generated. These split prompts correspond to each of the step-by-step processes A to C, from extracting a column object 206 from BIM data 205 of the building shown in the figure (process A), obtaining data on the surface of the extracted column object (process B), and obtaining the length 207 of the line connecting the surfaces of adjacent columns as the distance between the columns (process C).

[0042] Furthermore, the configuration may be such that the user can reconfigure the prompts to be set in the prompt setting unit based on the generated divided prompts.

[0043] By dividing the main prompt set earlier and generating split prompts in this way, it is possible to visualize the prompts (split prompts) that instruct the intermediate processes leading to a result, and evaluate the appropriateness of the processes and split prompts at each stage.

[0044] <Split Prompt Setting Unit> The split prompt setting unit 103 has a function of setting the generated split prompt. For example, the user can modify the generated split prompt. For example, in the above example, the split prompt "Please find the pillar" can be modified to "Please find the pillar. Note that "pillar" also means pillar."

[0045] The system may also be configured to allow further division of the split prompt. This division may be achieved by a user creating a further split prompt, or by inputting the split prompt into a trained AI model to generate further split prompts. By enabling corrections and adjustments on a split prompt basis in this way, it is possible to set and correct instructions for the trained AI model in detail, resulting in more accurate answers.

[0046] <Split processing step generation unit> The split processing step generation unit 104 has the function of inputting the generated split prompt or the set split prompt into the trained AI model and generating a split processing step according to the generated split prompt or the set split prompt.

[0047] A splitting process step is a specific process for executing the instructions of a splitting prompt. For example, a splitting process step corresponding to the splitting prompt "Find the pillars" illustrated in FIG. 2 may involve calling a trained AI model to extract pillar objects from a target BIM (Building Information Modeling) file or CAD (Computer Aided Design) file. Furthermore, a splitting process step corresponding to "Find the surface of each pillar" may involve calling a trained AI model to extract surface data of the pillar objects. Furthermore, a splitting process step corresponding to "Draw a line connecting the surfaces and find its length" may involve generating a script (a simple program) that obtains the length of the line connecting the surfaces of adjacent pillars and executing the script to obtain the distance between the pillars.

[0048] The generation of split processing steps as described above requires appropriately dividing the entire process and determining the processing to be executed for each divided unit, which has the problem that only engineers with specialized knowledge and experience can create them themselves. However, by generating split processing steps by inputting split prompts into a trained AI model, it is possible to obtain the effect that anyone can create split processing steps themselves. Furthermore, because the split prompts are also generated by a trained AI model, specialized knowledge, etc. is further unnecessary for generating split processing steps.

[0049] As described above, the division processing steps generated by the trained AI model may be processing by executing a script or processing via the trained AI. Furthermore, the division processing steps may use the processing results of the preceding and following division processing steps for processing, or the processing results of one division processing step may be provided to multiple other division processing steps. Furthermore, some or all of the division processing steps may be configured by referencing existing processing blocks for performing predetermined processing, or the division processing steps may be configured by modifying or adjusting existing processing blocks.

[0050] The split prompt generation unit can also be configured to input the set prompt into a trained AI model to generate split prompts by dividing the set prompt, and to generate split processing steps corresponding to each generated split prompt. As described above, the split processing step generation unit also generates split processing steps by inputting the split prompt into a trained AI model, so by configuring the split prompt generation unit to have this function as well, it is possible to generate corresponding split prompts and split processing steps together.

[0051] <Division Processing Step Result Acquisition Unit> The division processing step result acquisition unit 105 has a function of acquiring the processing results of the division processing step generated. For example, in the example shown in FIG. 2, the extracted pillars are the processing result first, and the surface data of the extracted pillars is the next processing result. Then, the length of the line connecting the surface data of each extracted pillar is acquired as the processing result. In this case, the final processing result of the division processing step becomes the answer to the prompt "Find the distance between pillars." Furthermore, the processing result of the division processing step may be further calculated, processed, etc., and the result may be configured to be the final answer to the prompt.

[0052] By obtaining the processing results of each segmentation process step, the correctness and validity of the intermediate process results leading to the final answer can be evaluated. If the result obtained in any of the segmentation process steps is not valid, the segmentation process step that produced that result and the segment prompt that generated that segmentation process step can be corrected or adjusted to make the final answer more appropriate.

[0053] <Sample setting unit> The sample setting unit has a function of setting correct answer sample file data that is input to the trained AI model together with the set prompt and serves as a correct answer example for obtaining the answer required by the set prompt, and correct answer information included in the correct answer sample file data.

[0054] The split prompt generation unit generates a split prompt by inputting the set correct answer sample file data and correct answer information into the trained AI model along with the set prompt. The split processing step generation unit generates a split processing step by inputting the set correct answer sample file data and correct answer information into the trained AI model. In this case, the correct answer sample file data and correct answer information can be set for each split prompt input into the trained AI model, or they can be set for some of the split prompts.

[0055] In addition, when the division processing step result acquisition unit acquires the processing results of the division processing step using a trained AI model, the results can be configured to be input into the trained AI model.

[0056] The correct sample file data is a sample file data used to obtain desired information through a set prompt, and contains correct information that serves as a correct example. For example, if the desired information is the illuminance distribution in a room, BIM data representing the room in which lighting and other equipment are installed would be the correct sample file data. Such correct sample file data and the lighting represented therein constitute the correct information.

[0057] In addition, not only is one piece of correct answer information set for correct answer sample file data, but also correct answer information may be set for each piece of correct answer sample file data. For example, if the correct answer sample file data is a 3D model file for multiple architectural structures including information on columns and beams, and the desired information is the "collision points between columns and beams" in each architectural structure, the actual number of "collision points between columns and beams" in each architectural structure is not limited to "1" and may vary, and the collision points between each column and beam are specified as correct answer information for each piece of correct answer sample file data. Note that the correct answer information in this case may be coordinate information on a 3D image showing the collision points, or may be image information that clarifies the collision points by, for example, showing them with a surrounding line.

[0058] The sample setting unit can also be configured to set the correct answer information by specifying specific information on the data on a viewer that displays the set correct answer sample file data. While the user's operation of specifying the correct answer information, such as a three-dimensional space model, two-dimensional vector data, image, table data, numerical values, or text, requires expertise, providing a viewer as in this configuration is beneficial because it makes it possible to easily set the correct answer information, such as a three-dimensional model, three-dimensional space, or two-dimensional graphic part, an image area, or a cell in table data.

[0059] 3 is a conceptual diagram showing an example of a viewer displaying correct sample file data. As shown in FIG. 3(a), BIM data representing a room in which lighting and other equipment are arranged is displayed as correct sample file data in the viewer 301. This BIM data includes various information about the dimensions and materials of the displayed room, as well as about objects such as each piece of equipment (ceiling light 302, bracket light 303, bench 304, counter 305), and each of these can be extracted.

[0060] Here, if the desired information is the illuminance distribution in a room and the ceiling light 302 is to be specified as the correct information from this correct sample file data, for example, as shown in FIG. 3( b), the ceiling light 302 can be designated as the correct information by selecting it by encircling it with a box 306. The bracket light 305 can also be similarly designated as the correct information. By designating the correct information in this way, information such as the illuminance and position of each of the ceiling light and bracket light is identified, and this becomes the correct information for obtaining the illuminance distribution in the room. Note that the correct information in this case can be designated not only by encircling it but also by directly selecting the 3D object.

[0061] The correct answer information can be set in various ways depending on the correct answer sample file data. For example, if the correct answer sample file data is an image file, by selecting a specific range in the viewer that opens the image file, that range can be set as the correct answer information to be extracted. Also, by selecting a specific layer in the viewer that opens a CAD file, that layer can be set as the correct answer information to be extracted.

[0062] It is also possible to set content that does not explicitly exist in the data. For example, by opening a BIM file in a viewer and selecting the space between specific objects in the viewer, the distance between those objects can be set as the correct information to be extracted. Also, by opening spreadsheet data such as Excel in the viewer and selecting a specific cell, that cell can be set as the correct information to be extracted.

[0063] As described above, the correct answer information to be set may be in any format, such as numerical information, a three-dimensional spatial model, a two-dimensional figure, an image, or text. Furthermore, the result of a calculation process may be set as the correct answer information. Furthermore, the correct answer information may not only provide information on the desired final result, but also information on the correct answer in the processing steps leading up to that result. For example, in the example of extracting "collision points between columns and beams" from the aforementioned three-dimensional model, the correct answer information in the processing steps may include (1) the column and beam models included in each file, (2) whether or not there is a collision between them, and (3) information on the three-dimensional space from which the collision points are extracted.

[0064] The sample setting unit can also be configured to set the correct sample file data as incorrect information by specifying specific information on the data in a viewer that displays the correct sample file data. In 3D models, there are often objects that are confusable with correct information. For example, if you want to extract an "air conditioning duct," objects like "water supply and drainage piping" and "wiring rack" may be placed near the correct object, which are similar objects. In this case, by specifying not only "air conditioning duct" as the correct answer but also "water supply and drainage piping" and "wiring rack" as incorrect answers, the trained AI can avoid excessive pipe detection and accurately obtain the desired information.

[0065] It is also possible that a file containing only incorrect answer information without correct answer information is set as correct answer sample file data.

[0066] <Correct Answer Information Explanation Prompt Setting Unit> The correct answer information explanation prompt setting unit has a function of setting a correct answer information explanation prompt that explains the set correct answer information. 3D models often contain objects that are confusable with the correct answer information object. For example, when extracting an "air conditioning duct," "water supply and drainage piping" and "wiring rack" may be placed near the object, and these are similar objects. In this case, instead of simply specifying "air conditioning duct" as the correct answer information, the user can specify "In this correct answer sample file data, this "air conditioning duct" is the correct answer for the desired information." This is because they are long, thin cylindrical models and their ends are connected to "air intakes." " By setting a prompt that explains the correct answer information, it is possible to prevent the trained AI model from detecting incorrect information, generate accurate segmentation prompts and segmentation processing steps, and improve the accuracy of the results obtained when obtaining the processing results of the segmentation processing steps using the trained AI model. Furthermore, if multiple pairs of correct sample file data and correct answer information are input as samples to the trained AI model, a correct answer information explanation prompt may be set for each pair.

[0067] Furthermore, if incorrect information is specified, an explanation for this incorrect information may be included in the correct information explanation prompt.

[0068] Furthermore, the correct answer sample file data and correct answer information may be set by the user, or may be set in advance by a person who provides the information processing device to the user.

[0069] In addition, the split prompt generation unit can be configured to input the set correct answer sample file data and correct answer information together with the set prompt into a trained AI model to generate a split prompt.

[0070] For example, by using the BIM file shown in Figure 2 as correct sample file data and setting the column object in the figure as correct answer information, it is possible to contribute to the generation of a more appropriate split prompt such as, "Please refer to the column object specified in the correct answer information and find the column. Also, please note that the specified vertical object is not a column but a piping duct." Also, by selecting the space between adjacent column objects and setting the distance between those column objects as correct answer information to be extracted, it is possible to generate a split prompt for appropriately extracting the information necessary to measure the distance between the columns.

[0071] By inputting the correct sample file data and correct answer information into the trained AI model along with the prompt, an appropriate splitting prompt can be generated to obtain the desired information, and the splitting processing step generated based on the splitting prompt generated in this way will also be a more appropriate splitting processing step for obtaining the desired information.

[0072] The split processing step generation unit can also be configured to input the set correct answer sample file data and correct answer information into a trained AI model to generate split processing steps. As described above, instead of inputting correct answer information, etc. when generating split prompts, it is also effective to input correct answer information, etc. when generating split processing steps. Correct answer information, etc. may also be input at both the split prompt generation stage and the split processing step generation stage. The correct answer information, etc. to be input may be the same, or different correct answer information, etc. may be input at each stage. Correct answer information, etc. may also be set for each split prompt, and the split processing steps may be generated by inputting them. Furthermore, when a trained AI model is used to obtain the processing results of the split processing steps, the correct answer information, etc. may also be input into the trained AI model to obtain the processing results.

[0073] <Split processing step evaluation unit> The system can also be configured to further include a split processing step evaluation unit that compares the processing results of a split processing step generated by inputting correct answer sample file data and correct answer information set in the trained AI model, or a split processing step generated from a split prompt generated by inputting correct answer sample file data and correct answer information set in the trained AI model, with the correct answer information, and evaluates the generated split processing step.

[0074] The evaluation of the division processing steps may be performed on the processing result of the final division processing step among the generated division processing steps, or on the processing results of all the generated division processing steps. By comparing the processing result with the correct answer information in this way, it is possible to evaluate how accurate the division processing steps are.

[0075] Furthermore, by configuring the segmentation processing step generation unit to further input the evaluation results into the trained AI model, segmentation processing steps can be generated based on the evaluation results, which can be useful for generating more accurate segmentation processing steps. For example, it can be configured to continue regenerating segmentation processing steps until the processing results and correct answer information match.

[0076] <Hardware Configuration> Fig. 4 is a conceptual diagram showing an example of the configuration of hardware that realizes the information processing apparatus of Example 1. As shown in the figure, the information processing apparatus 400 has a CPU 401 that performs various arithmetic processing, a RAM 402 that is a volatile storage medium, a storage 403 such as a flash memory or HDD that is a non-volatile storage medium, a communication interface 404, and an input / output interface 405. The RAM 402 reads programs that perform various arithmetic processing to cause the CPU 401 to execute them, and also provides a work area (working area) for the programs. In addition, multiple addresses are assigned to the RAM 402, and the programs executed by the CPU 401 can exchange data and perform processing by identifying and accessing these addresses.

[0077] Here, the functions of the prompt setting unit 101, the split prompt generation unit 102, the split prompt setting unit 103, the split processing step generation unit 104, and the split processing step result acquisition unit 105 of the information processing device 100 in Figure 1 are mainly realized by the CPU 401 and RAM 402 in Figure 4. Also, the functions of the sample setting unit, the correct answer information explanation prompt setting unit, and the split processing step evaluation unit (not shown) are mainly realized by the CPU 401 and RAM 402 in Figure 4. Also, when an external database that stores file data to be used for obtaining desired information or setting correct answer sample file data exists, or when an external trained AI model is used, each function is realized by mutually exchanging signals and information via the communication interface 404 and the input / output interface 405.

[0078] <Processing Flow> FIG. 5 is a flow diagram illustrating a simplified example of the processing flow of the information processing device according to the first embodiment. First, a prompt is set (S501: prompt setting step). Then, the set prompt is input to a trained AI model to generate split prompts by splitting the set prompt (S502: split prompt generation step). Then, the generated split prompts are set (S503: split prompt setting step). Then, the generated split prompts or the set split prompts are input to the trained AI model to generate split processing steps corresponding to the generated split prompts or the set split prompts (S504: split processing step generation step). Then, the processing results of the generated split processing steps are obtained (S505: split processing step result acquisition step).

[0079] <Effects> According to the information processing apparatus of this embodiment, it is possible to evaluate and correct the processes along the way to the final result.

[0080] Second Embodiment Overview In contrast to the information processing apparatus of the first embodiment, which generates divided processing steps from divided prompts, this embodiment is an information processing apparatus that generates divided processing steps (processing steps) from prompts without going through divided prompts.

[0081] <Functional Configuration> Fig. 6 is a block diagram showing an example of the functional configuration of the information processing device of this embodiment. As shown in Fig. 6, the information processing device 600 has a prompt setting unit 601, a processing step generation unit 602, a processing step result acquisition unit 603, and a processing step prompt generation unit 604. The prompt setting unit has the same function as the component with the same name in Example 1, so a description thereof will be omitted.

[0082] <Processing Step Generation Unit> The processing step generation unit 602 has a function of inputting the set prompt to the trained AI model and generating divided processing steps based on the set prompt. In Example 1, the divided prompt was input to the trained AI model to generate divided processing steps, but in this configuration, the set prompt is input to generate processing steps.

[0083] The reason for calling it a "processing step" is that the prompt from which it is generated is different from that in Example 1, but the technical significance is the same as that of the "divided processing step," and the trained AI model infers the multiple processes included in the prompt to generate individual processing steps. The generated steps may be processes using a static script or processes via the trained AI model. Furthermore, since the generated processing steps are the same as the divided processing steps in Example 1, a repeated explanation will be omitted.

[0084] <Processing Step Result Acquisition Unit> The processing step result acquisition unit 603 has a function of acquiring the processing results of the generated processing steps. The function of this configuration is the same as that of the divided processing step result acquisition unit in the first embodiment. By acquiring the processing results of each processing step, the correctness and validity of the results of the intermediate processing can be evaluated. If the result obtained in any processing step is inappropriate, the processing step that brought about that result and the prompt that generated that processing step can be corrected or adjusted, thereby making the final answer more appropriate.

[0085] <Processing step prompt generation unit> The processing step prompt generation unit 604 has the function of inputting the generated processing step into the trained AI model and generating a processing step prompt, which is a prompt that instructs the trained AI model to use the generated processing step.

[0086] The processing step prompt that instructs the trained AI model to execute the generated processing step is a prompt that the generated processing step is generated by the trained AI model, and is a prompt that corresponds to the processing step. By generating such a processing step prompt, it is possible to evaluate whether each generated processing step and its corresponding processing step prompt are correctly associated, which can contribute to obtaining a more accurate answer.

[0087] Furthermore, similar to the split prompt setting unit of Example 1, it can also be configured to have a function that allows a user or the like to set the generated processing step prompt, and the user can modify or adjust the generated processing step prompt.

[0088] The generated processing step prompt may be input to a trained AI model to generate further processing steps. By obtaining the processing results of the regenerated processing steps and repeating the generation of processing step prompts and processing steps based on the obtained results, it is possible to obtain a more accurate answer.

[0089] <Other configurations> As in Example 1, when using a trained AI model to generate processing steps or processing step prompts or to obtain processing results of processing steps, it is also possible to configure the trained AI model to input correct sample file data and correct answer information (sample setting unit).

[0090] The system can also be configured to compare the acquired processing results with the set correct answer sample file data and correct answer information to evaluate the processing steps (processing step evaluation unit). It can also be configured to set a correct answer information explanation prompt that explains the set correct answer information (correct answer information explanation prompt setting unit). In this way, using the correct answer sample file and correct answer information helps generate more appropriate processing steps and obtain the desired answer with high accuracy.

[0091] <Hardware Configuration> The information processing apparatus of this embodiment can be realized by the hardware configuration shown in Fig. 4 according to the first embodiment. The functions of a prompt setting unit 601, a processing step generation unit 602, a processing step result acquisition unit 603, and a processing step prompt generation unit 604 of an information processing apparatus 600 of the second embodiment shown in Fig. 6 are mainly realized by the CPU 401 and RAM 402 of Fig. 4.

[0092] <Processing Flow> FIG. 7 is a flow diagram illustrating a simplified example of the processing flow of the information processing device according to the second embodiment. First, a prompt is set (S701: prompt setting step). Then, the set prompt is input to the trained AI model, and processing steps divided based on the set prompt are generated (S702: processing step generation step). Then, a determination is made as to whether or not to generate a processing step prompt (S703: determination step). If a determination is made to generate a processing step prompt, the generated processing step is input to the trained AI model, and a processing step prompt that instructs the trained AI model to use the generated processing step is generated (S704: processing step prompt generation step). Furthermore, a processing step is generated again (S702). On the other hand, if a determination is made not to generate a processing step prompt, the processing result of the generated processing step is acquired (S705: processing step result acquisition step).

[0093] <Effects> According to the information processing apparatus of this embodiment, it is possible to evaluate and correct the processes along the way to the final result.

[0094] 100, 400, 600: Information processing device 101, 601: Prompt setting unit 102: Divided prompt generation unit 103: Divided prompt setting unit 104: Divided processing step generation unit 105: Divided processing step result acquisition unit 602: Processing step generation unit 603: Processing step result acquisition unit 604: Processing step prompt generation unit 401: CPU 402: RAM 403: Storage 404: Communication interface 405: Input / output interface

Claims

1. An information processing device having: a prompt setting unit that sets a prompt; and a split prompt generation unit that inputs the set prompt into a trained AI model and generates split prompts by dividing the set prompt.

2. The information processing apparatus according to claim 1, further comprising a divided prompt setting unit that sets the generated divided prompt.

3. The information processing device according to claim 2, further comprising a division processing step generation unit that inputs the generated division prompt or the set division prompt into the trained AI model and generates a division processing step corresponding to the generated division prompt or the set division prompt.

4. The information processing device according to claim 3, further comprising a division processing step result acquisition unit that acquires the processing results of the generated division processing steps.

5. An information processing device as described in claim 1 or 2, wherein the split prompt generation unit inputs the set prompt into the trained AI model and generates split prompts by splitting the set prompt, together with the corresponding split processing steps.

6. An information processing device according to claim 1, further comprising a sample setting unit that sets correct answer sample file data that serves as a correct answer example for obtaining the answer required by the set prompt, and correct answer information contained in the correct answer sample file data.

7. The information processing device according to claim 6, further comprising a correct answer information explanation prompt setting unit that sets a correct answer information explanation prompt that explains the correct answer information that has been set.

8. The information processing device according to claim 6, wherein the split prompt generation unit inputs the set correct sample file data and correct answer information together with the set prompt into a trained AI model to generate a split prompt.

9. The information processing device according to claim 8, wherein the split processing step generation unit inputs the set correct sample file data and correct answer information into the trained AI model to generate split processing steps.

10. An information processing device as described in claim 9, further comprising a split processing step evaluation unit that compares the processing result of the split processing step generated by inputting the set correct answer sample file data and correct answer information into the trained AI model, or the split processing step generated from a split prompt generated by inputting the set correct answer sample file data and correct answer information into the trained AI model, with the correct answer information, to evaluate the generated split processing step.

11. An information processing device having: a prompt setting unit that sets a prompt; and a processing step generation unit that inputs the set prompt into the trained AI model and generates divided processing steps based on the set prompt.

12. The information processing device according to claim 11, further comprising a processing step result acquisition unit that acquires the processing results of the generated processing steps.

13. An information processing device as described in claim 11 or 12, further comprising a processing step prompt generation unit that inputs the generated processing step into a trained AI model and generates a processing step prompt that instructs the trained AI model to use the generated processing step.

14. An information processing device according to claim 13, wherein the processing step generation unit inputs the generated processing step prompt into the trained AI model to generate a processing step corresponding to the processing step prompt.

15. An information processing method executed by an information processing device, comprising: a prompt setting step of setting a prompt; and a split prompt generation step of inputting the set prompt into a trained AI model and generating split prompts by dividing the set prompt.

16. An information processing program that causes an information processing device to execute: a prompt setting step for setting a prompt; and a split prompt generation step for inputting the set prompt into a trained AI model and generating split prompts by dividing the set prompt.

17. An information processing method executed by an information processing device, comprising: a prompt setting step for setting a prompt; and a processing step generation step for inputting the set prompt into the trained AI model and generating divided processing steps based on the set prompt.

18. An information processing program that causes an information processing device to execute: a prompt setting step for setting a prompt; and a processing step generation step for inputting the set prompt into the trained AI model and generating divided processing steps based on the set prompt.

Images