Information processing device, information processing method, and program

Abstract

By combining and utilizing multiple generative AIs in a more optimal manner, the accuracy of the generative AI's responses can be further improved. [Solution] An information processing device comprising: a first interpretation processing unit that, upon input to a first generating AI, provides input data including character information and a first prompt including a first instruction relating to the interpretation of the content of the input data, causing the first generating AI to output first output data including the result of interpreting the content of the character information; and a second interpretation processing unit that, upon input to a second generating AI, provides input data, first output data, and a second prompt including at least a first instruction and a second instruction relating to the evaluation of the output of the first generating AI, causes the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI included in the first output data, and outputs second output data including the result of interpreting the content of the character information reflecting the evaluation result.

Description

Technical Field

[0004] , , , , , , ,

[0003] , ,

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] With the development of large language model (LLM) technology, various multi-modal LLMs that can handle various types of information such as audio and images as well as text have been proposed. For example, some multi-modal LLMs that can handle images can perform character recognition from images and answer instructions input to prompts without separately using OCR (Optical Character Recognition). Examples of multi-modal LLMs that can handle images include Gemini (registered trademark) provided by Google LLC and Claude (registered trademark) provided by Anthropic. Thus, in recent years, the scope of application of LLMs has been expanding.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In recent years, techniques for using a combination of multiple LLMs have also been studied. For example, Patent Document 1 discloses a technique for assisting in the creation of test cases by combining multiple LLMs. On the other hand, in the technique disclosed in Patent Document 1, it is necessary to separately train each of the multiple LLMs to be specialized for a specific task, which is laborious and costly to introduce and has limited applications. Against this backdrop, research is being conducted on technologies that allow the use of multiple LLMs (hereinafter also referred to as generative AI) in combination without requiring significant effort or cost.

[0005] Therefore, this disclosure proposes a technology to further improve the accuracy of the responses of a generative AI by combining and using multiple generative AIs in a more suitable manner. [Means for solving the problem]

[0006] The information processing device according to the present invention includes: a first interpretation processing unit that, upon inputting input data including character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data to a first generating AI, causes the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data; and a second interpretation processing unit that, upon inputting the input data, the first output data, and a second prompt including at least a first instruction relating to the evaluation of the output of the first generating AI to a second generating AI, causes the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI included in the first output data, and outputs second output data including the result of interpreting the content of the character information contained in the input data, reflecting the evaluation result. [Effects of the Invention]

[0007] According to the present invention, by combining and using multiple generative AIs in a more suitable manner, it becomes possible to further improve the accuracy of the generative AI's responses. [Brief explanation of the drawing]

[0008] [Figure 1] This is a diagram illustrating the general functions of an information processing device. [Figure 2] This diagram shows an example of the hardware configuration of an information processing device. [Figure 3] This is a functional block diagram showing an example of the functional configuration of an information processing device. [Figure 4] It is a flowchart showing an example of the processing of an information processing apparatus. [Figure 5] It is a diagram showing an example of input data. [Figure 6] It is a diagram showing an example of a prompt. [Figure 7] It is a diagram showing an example of output data. [Figure 8] It is a diagram showing an example of input data. [Figure 9] It is a diagram showing an example of output data. [Figure 10] It is a diagram showing an example of input data. [Figure 11] It is a diagram showing an example of output data. [Figure 12] It is a diagram showing an example of output data. [Figure 13] It is a diagram showing an example of output data. [Figure 14] It is a diagram showing an example of output data. [Figure 15] It is a diagram showing an example of a prompt. [ [Figure 16] It is a diagram showing an example of output data. [Figure 17] It is a diagram showing an example of output data. [Figure 18] It is a diagram showing an example of output data. [[ID=4~5]] [Figure 19] It is a diagram showing an example of output data. [Figure 20] It is a diagram showing an example of output data. [Figure 21] It is a diagram showing an example of output data.

Embodiments for Carrying Out the Invention

[0009] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

[0010] <Functional Overview> Referring to FIG. 1, the functional overview of the information processing apparatus according to an embodiment of the present disclosure will be described. The information processing apparatus 100 according to this embodiment uses a learning model such as a neural network that has been trained based on deep learning or machine learning, particularly a model called Generative AI, to execute a task instructed by a user and output the result. The information processing apparatus 100 according to this embodiment combines multiple generative AIs (LLMs) that perform substantially the same functions, such as Gemini and Claude, in multiple stages to improve the accuracy of the output results compared to using a single generative AI. In this embodiment, for the sake of convenience, various explanations will be made assuming that the generative AI interprets the content of a manuscript including character information such as a document (for example, the content indicated by the character information) and outputs data indicated by the interpretation result in character information.

[0011] In the example shown in FIG. 1, the first generative AI 510 and the second generative AI 520 that perform substantially the same functions are applied. Each of the first generative AI 510 and the second generative AI 520 receives as input data at least data including character information such as a document, and a generative AI that has been trained to interpret the content indicated by the data (at least the content indicated by the character information included in the data) according to the instructions shown in the prompt is applied. Also, the first generative AI 510 and the second generative AI 520 can apply a so-called multimodal LLM that can receive various types of data such as image data, not limited to only so-called text data, as input data. Examples of such generative AIs include Gemini, Claude, etc. as described above, and various other multimodal LLMs have been publicly disclosed. In the example shown in FIG. 1, it is assumed that the input data D100 is image data generated by reading a paper manuscript with a scanner.

[0012] In the example shown in Figure 1, the information processing device 100 first causes the first generating AI 510 to interpret the content indicated by the input data D100 (the content indicated by the character information included in the input data D100 as content such as a document). Specifically, the information processing device 100 inputs input data D100 and a first prompt D201 which includes at least instructions relating to the interpretation of the content indicated by the character information contained in the input data to the first generating AI 510. In this way, the information processing device 100 causes the first generating AI 510 to interpret the content indicated by the character information contained in the input data D100 and to output first output data D301 which includes at least the result of said interpretation. The specification of input data D100 and the instructions regarding the interpretation of the content of said input data D100 included in the first prompt D201 may be accepted, for example, as user input.

[0013] Next, the information processing device 100 causes the second generating AI 520 to interpret the content indicated by the input data D100 (the content indicated by the character information contained in the input data D100). At this time, the information processing device 100 inputs output data D301 to the second generating AI 520, causing it to evaluate the validity of the interpretation result of the content indicated by the input data D100 by the first generating AI 510, and to take the result of this evaluation into consideration when interpreting the content indicated by the input data D100. Specifically, the information processing device 100 generates a second prompt D202 as a prompt to be input to the second generation AI 520, which includes instructions relating to the interpretation of the content indicated by the character information contained in the input data (similar to the instructions included in the first prompt D201) and instructions relating to the evaluation of the result of the interpretation by the other generation AI. Then, the information processing device 100 inputs the input data D100, the first output data D301 output by the first generation AI 510, and the second prompt D202 to the second generation AI 520. In this way, the information processing device 100 causes the second generation AI 520 to evaluate the validity of the interpretation result of the content indicated by the input data D100 by the first generation AI 510, then to interpret the content indicated by the input data D100 taking the evaluation result into account, and to output a second output data D302 that includes at least the result of the interpretation. Furthermore, at this time, the information processing device 100 may cause the second generating AI 520 to output the validity evaluation result. The validity evaluation result may be included in the second output data D302, or the second output data D302 may be output as separate data.

[0014] As described above, the information processing device 100 according to this embodiment has multiple generating AIs interpret the content of the input data in a stepwise manner. In this process, the information processing device 100 has the subsequent generating AI evaluate the interpretation result of the preceding generating AI regarding the content of the input data, and feeds this evaluation result back to the subsequent generating AI when interpreting the content of the input data. With this configuration, it is expected that the accuracy of the interpretation will be improved compared to when each generating AI individually interprets the content of the input data. The details of this effect, along with specific application examples, will be described separately in an example.

[0015] <Hardware Configuration> Referring to Figure 2, an example of the hardware configuration of an information processing device 900 applicable as an apparatus for realizing the technology according to one embodiment of the present disclosure (for example, the information processing device 100 described above) will be explained. As shown in Figure 2, the information processing device 900 according to this embodiment includes a CPU (Central Processing Unit) 910, a ROM (Read Only Memory) 920, and a RAM (Random Access Memory) 930. The information processing device 900 also includes an auxiliary storage device 940 and a network I / F 970. The information processing device 900 may also include at least one of an output device 950 and an input device 960. The CPU 910, ROM 920, RAM 930, auxiliary storage device 940, output device 950, input device 960, and network I / F 970 are interconnected via a bus 980.

[0016] The CPU 910 is a central processing unit that controls various operations of the information processing unit 900. For example, the CPU 910 may control the operation of the entire information processing unit 900. The ROM 920 stores control programs, boot programs, and other programs that can be executed by the CPU 910. The RAM 930 is the main memory of the CPU 910 and is used as a work area or a temporary storage area for deploying various programs.

[0017] The auxiliary storage device 940 stores various data and programs. The auxiliary storage device 940 is implemented by a storage device capable of temporarily or permanently storing various data, such as an HDD (Hard Disk Drive) or non-volatile memory such as an SSD (Solid State Drive).

[0018] The output device 950 is a device that outputs various types of information and is used to present various types of information to the user. For example, the output device 950 may be implemented by a display device such as a display. In this case, the output device 950 presents information to the user by displaying various types of display information. As another example, the output device 950 may be implemented by an acoustic output device that outputs sounds such as voice or electronic sounds. In this case, the output device 950 presents information to the user by outputting sounds such as voice or electronic sounds. Furthermore, the device to which the output device 950 is applied may be appropriately changed depending on the medium used to present information to the user.

[0019] The input device 960 is used to receive various instructions from the user. In this embodiment, the input device 960 includes input devices such as a mouse, keyboard, and touch panel. As another example, the input device 960 may also include a sound collection device such as a microphone to collect the voice spoken by the user. In this case, various analysis processes such as acoustic analysis and natural language processing are applied to the collected voice so that the content of the voice is recognized as an instruction from the user. Furthermore, the device applied as the input device 960 may be changed as appropriate depending on the method of recognizing the user's instructions. In addition, multiple types of devices may be applied as the input device 960.

[0020] The network interface 970 is used for communication with external devices via a network. The device used as the network interface 970 may be changed as appropriate depending on the type of communication path and the applicable communication method.

[0021] The CPU 910 loads the program stored in the ROM 920 or auxiliary storage device 940 into the RAM 930 and executes this program, thereby realizing the functional configuration of the information processing device shown in Figure 3, and the processing of the information processing device described later with reference to Figure 4.

[0022] The above describes an example of the hardware configuration of an information processing device 900 that can be used as a device constituting the information processing system 1 according to this embodiment, with reference to Figure 2.

[0023] <Functional Configuration> Referring to Figure 3, an example of the functional configuration of the information processing device 100 according to this embodiment will be described. Note that the generation AI 510 and 520 shown in Figure 3 correspond to the first generation AI 510 and second generation AI 520 shown in Figure 1. Also, the input data D100 and output data D302 correspond to the input data D100 and second output data D302 shown in Figure 1. As shown in Figure 1, the information processing device 100 according to this embodiment includes an input receiving unit 101, a prompt generation unit 102, a first interpretation processing unit 103, and a second interpretation processing unit 104.

[0024] The input receiving unit 101 schematically shows an input interface that receives the specification of input data D100 (input data subject to content interpretation processing) to be processed by the generating AI, and instructions for processing the input data D100 (instructions related to the interpretation of the content of the input data). As mentioned above with reference to Figure 1, in this embodiment, the input data D100 is data that includes at least character information such as a document, and the above instructions include instructions related to the interpretation of the content of the character information contained in the input data D100.

[0025] The prompt generation unit 102 generates prompts that will be inputs to the first generation AI 510 and the second generation AI 520, respectively, based on the instructions received as input by the input reception unit 101, that is, instructions for processing the specified input data. Specifically, the prompt generation unit 102 generates a first prompt D201, which will be input to the first generation AI 510, such that it includes at least the instructions received as input by the input reception unit 101 (in particular, instructions relating to the interpretation of the content of the input data, hereinafter also referred to as the first instructions). Furthermore, the prompt generation unit 102 generates a second prompt D202, which will be input to the second generation AI 520, such that it includes at least the instruction received by the input reception unit 101 as input (in particular, the first instruction) and an instruction related to the evaluation of the validity of the interpretation result of the input data based on the instruction of the first prompt D201 by the first generation AI (hereinafter also referred to as the second instruction). Thus, the first prompt D201 and the second prompt D202 both contain the same instruction (in particular, the first instruction) that the input receiving unit 101 has received as input. Furthermore, the second prompt D202 also contains the second instruction. The contents of the first prompt D201 and the second prompt D202 will be described separately later with specific examples as embodiments.

[0026] The first interpretation processing unit 103 acquires the input data D100 that the input receiving unit 101 has designated as input, and the first prompt D201 generated by the prompt generation unit 102, and uses these data as input to the first generation AI 510. In this way, the first interpretation processing unit 103 causes the first generation AI 510 to interpret the content indicated by the input data D100 (the content indicated by the character information contained in the input data D100) based on the first instruction contained in the first prompt D201. The first interpretation processing unit 103 then acquires output data D301 from the first generation AI 510 that includes at least the interpretation result of the input data D100 in accordance with the instruction shown in the first prompt D201.

[0027] The second interpretation processing unit 104 acquires the input data D100 that the input receiving unit 101 has designated as input, the second prompt D202 generated by the prompt generation unit 102, and the first output data D301 which includes the interpretation result of the content of the input data D100 by the first generation AI 510, and uses these data as input to the second generation AI 520. In this way, the second interpretation processing unit 104 causes the second generation AI 520 to evaluate the validity of the interpretation result of the content of the input data D100 by the first generation AI 510 (the validity of the interpretation result of the content indicated by the character information contained in the input data D100) based on the second instruction contained in the second prompt D202. Then, the second interpretation processing unit 104 causes the second generation AI 520 to interpret the content indicated by the input data D100, reflecting the result of the validity evaluation, based on the first instruction contained in the second prompt D202. Then, the second interpretation processing unit 104 obtains output data D302 from the second generation AI 520, which includes at least the interpretation result of the input data D100 based on the second prompt D202. Furthermore, the second interpretation processing unit 104 may obtain information from the second generation AI 520 indicating the evaluation result of the validity of the interpretation result of the content of the input data D100 by the first generation AI 510. This information indicating the validity evaluation result may be included in the output data D302, or it may be obtained as separate data from the output data D302. The second interpretation processing unit 104 then outputs the acquired output data D302 to a predetermined output destination. As a specific example, the second interpretation processing unit 104 may present the information contained in the output data D302 to the user by displaying it in a predetermined display area (for example, the display of a terminal used by the user).

[0028] It should be noted that the configuration shown in Figure 3 is merely an example and does not limit the functional configuration of the information processing device 100 according to this embodiment. For example, the functions of each of the series of components shown in Figure 3 may be realized by the cooperation of multiple devices. As a specific example, the functions of some of the components and other components shown in Figure 3 may be realized by different devices connected by a network or the like. As another example, the processing load of some of the components may be distributed among multiple devices.

[0029] An example of the functional configuration of the information processing device 100 according to this embodiment has been described above with reference to Figure 3.

[0030] <Processing> Referring to Figure 4, an example of the processing performed by the information processing device 100 according to this embodiment will be described.

[0031] In S101, the input receiving unit 101 receives the specification of the input data D100 to be processed by the generated AI, and instructions for processing the input data D100. In S102, the prompt generation unit 102 generates a first prompt and a second prompt, which are inputs to the first generation AI 510 and the second generation AI 520, based on the processing instructions for the input data D100 received by the input reception unit 101 in S101. As mentioned above, the first prompt includes at least the instructions received by the input reception unit 101 as input (in particular, the first instruction relating to the interpretation of the content of the input data). The second prompt also includes the instructions received by the input reception unit 101 as input (in particular, the first instruction mentioned above), similar to the first prompt D201, and further includes a second instruction relating to the evaluation of the validity of the interpretation result of the input data based on the instructions of the first prompt D201 by the first generation AI.

[0032] In S103, the first interpretation processing unit 103 takes the input data D100 specified as input in S101 and the first prompt D201 generated in S102 as inputs to the first generation AI 510. In this way, the first interpretation processing unit 103 causes the first generation AI 510 to interpret the content indicated by the input data D100 based on the first instruction contained in the first prompt D201. The first interpretation processing unit 103 then obtains output data D301 from the first generation AI 510, which includes at least the interpretation result of the input data D100 corresponding to the instruction shown in the first prompt D201.

[0033] In S104, the second interpretation processing unit 104 takes the input data D100 specified as input in S101, the second prompt D202 generated in S102, and the output data D301 acquired in S103 as inputs to the second generation AI 520. In this way, the second interpretation processing unit 104 causes the second generation AI 520 to evaluate the validity of the interpretation result of the content of the input data D100 by the first generation AI 510, based on the second instruction included in the second prompt D202. Then, the second interpretation processing unit 104 causes the second generation AI 520 to interpret the content indicated by the input data D100, reflecting the result of the validity evaluation, based on the first instruction included in the second prompt D202. The second interpretation processing unit 104 then obtains output data D302 from the second generation AI 520, which includes at least the interpretation result of the input data D100 based on the second prompt D202. The second interpretation processing unit 104 may also obtain information from the second generation AI 520 indicating the evaluation result of the validity of the interpretation result of the content of the input data D100 by the first generation AI 510. In S105, the second interpretation processing unit 104 outputs the output data D302 acquired in S104 to a predetermined output destination.

[0034] An example of the processing of the information processing device 100 according to this embodiment has been described above with reference to Figure 4.

[0035] <Examples> An embodiment of the information processing device 100 according to one embodiment of this disclosure will be described below with specific examples.

[0036] (Example 1) An example 1 of an information processing device according to one embodiment of the present disclosure will be described with reference to Figures 5 to 11. In this embodiment, we will describe an example in which, using image data corresponding to the result of scanning a document as input data, Gemini provided by Google LLC is applied as the first generation AI 510 and Claude provided by Anthropic is applied as the second generation AI 520 in the configuration shown in Figure 1.

[0037] The input data D101 shown in Figure 5 corresponds to an example of data that would be applied as input data D100 in the example shown in Figure 1. Figure 6 shows an example of a prompt. Specifically, the first prompt D211 shown in Figure 6(a) corresponds to an example of the first prompt D201 in the example shown in Figure 1, and includes the first instruction D2111 related to the interpretation of the content indicated by the input data (input data D101, and input data D102 and D103, which will be described separately). Furthermore, the second prompt D212 shown in Figure 6(b) corresponds to an example of the second prompt D202 in the example shown in Figure 1, and includes a first instruction D2121 relating to the interpretation of the content indicated by the input data, and a second instruction D2122 relating to the evaluation of the validity of the interpretation result of the content indicated by the input data by the preceding generation AI (first generation AI 510). Note that the first instruction D2121 included in the second prompt D212 is substantially the same as the first instruction D2111 included in the first prompt D211.

[0038] Figure 7 shows an example of output data. Specifically, the first output data D311 shown in Figure 7(a) is an example of output data from the first generating AI 510, which takes the input data D101 shown in Figure 5 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D311 includes the interpretation result D3111 of the content indicated by the input data D101 by the first generating AI 510. Furthermore, the second output data D312 shown in Figure 7(b) is an example of output data from the second generating AI 520, which takes the input data D101 shown in Figure 5, the second prompt D212 shown in Figure 6(b), and the first output data D311 shown in Figure 7(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D312 includes the interpretation result D3121 of the content indicated by the input data D101 by the second generating AI 520, and the evaluation result D3122 of the validity of the interpretation result D3111 of the content indicated by the input data D101 by the first generating AI 510. As can be seen by referring to the evaluation result D3122 included in the second output data D312, the second generating AI 520 points out an error in the interpretation result of the input data D101 by the first generating AI 510. Furthermore, as can be seen by comparing the input data D101, the interpretation result D3111 included in the first output data D311, and the interpretation result D3112 included in the second output data D312, the accuracy of the interpretation of the content of the input data D101 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0039] Furthermore, the input data D102 shown in Figure 8 corresponds to an example of data that would be applied as input data D100 in the example shown in Figure 1. In contrast, Figure 9 shows an example of output data. Specifically, the first output data D321 shown in Figure 9(a) is an example of output data from the first generating AI 510, which takes the input data D102 shown in Figure 8 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D321 includes the interpretation result D3211 of the content indicated by the input data D102 by the first generating AI 510. Furthermore, the second output data D322 shown in Figure 9(b) is an example of output data from the second generating AI 520, which takes the input data D102 shown in Figure 8, the second prompt D212 shown in Figure 6(b), and the first output data D321 shown in Figure 9(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D322 includes the interpretation result D3221 of the content indicated by the input data D102 by the second generating AI 520, and the evaluation result D3222 of the validity of the interpretation result D3211 of the content indicated by the input data D102 by the first generating AI 510. As can be seen by referring to the evaluation result D3222 included in the second output data D322, the second generating AI 520 points out an error in the interpretation result of the first generating AI 510 regarding the content indicated by the input data D102. Furthermore, as can be seen by comparing the input data D102, the interpretation result D3211 included in the first output data D321, and the interpretation result D3212 included in the second output data D322, the accuracy of the interpretation of the content indicated by the input data D102 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0040] Furthermore, the input data D103 shown in Figure 10 corresponds to an example of data that would be applied as input data D100 in the example shown in Figure 1. In contrast, Figure 11 shows an example of output data. Specifically, the first output data D331 shown in Figure 11(a) is an example of output data from the first generating AI 510, which takes the input data D103 shown in Figure 10 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D331 includes the interpretation result D3311 of the content indicated by the input data D102 by the first generating AI 510. Furthermore, the second output data D332 shown in Figure 11(b) is an example of output data from the second generating AI 520, which takes the input data D103 shown in Figure 10, the second prompt D212 shown in Figure 6(b), and the first output data D331 shown in Figure 11(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D332 includes the interpretation result D3321 of the content indicated by the input data D103 by the second generating AI 520, and the evaluation result D3322 of the validity of the interpretation result D3311 of the content indicated by the input data D103 by the first generating AI 510. As can be seen by referring to the evaluation result D3322 included in the second output data D332, the second generating AI 520 points out an error in the interpretation result of the first generating AI 510 regarding the content indicated by the input data D103. Furthermore, as can be seen by comparing the input data D103, the interpretation result D3311 included in the first output data D331, and the interpretation result D3312 included in the second output data D332, the accuracy of the interpretation of the content indicated by the input data D103 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0041] The above describes an embodiment of the information processing device according to one embodiment of the present disclosure, shown in Figures 5 to 11. As illustrated in Figures 7, 9, and 11, it is conceivable that the accuracy of the interpretation result of the input data content by the generating AI may vary depending on the content of the input data. However, in all of the cases described above as embodiments, the accuracy of the interpretation of the input data content is improved by using multiple generating AIs in combination, as illustrated in Figure 1, compared to using a generating AI alone. From this, it can be expected that the information processing device according to this embodiment will have the effect of improving the accuracy of the interpretation of the input data content compared to using a generating AI alone.

[0042] (Example 2) Referring to Figures 12 to 14, an embodiment 2 of the information processing device according to one embodiment of the present disclosure will be described. In this embodiment, an example will be described in which a different combination from that of the first generation AI 510 and the second generation AI 520 is applied. Specifically, this embodiment describes an example in which, in the configuration shown in Figure 1, GPT-4o provided by OpenAI, Inc. is applied as the first generation AI 510, and Gemini provided by Google LLC is applied as the second generation AI 520. The input data will be the same as in Embodiment 1, with input data D101 shown in Figure 5, input data D102 shown in Figure 8, and input data D103 shown in Figure 10 being applied.

[0043] Figure 12 shows an example of the output data for the first generation AI510 and the second generation AI520 when the input data D101 shown in Figure 5 is applied. Specifically, the first output data D313 shown in Figure 12(a) is an example of the output data of the first generating AI 510, which takes input data D101 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to the example of the first output data D301 shown in Figure 1. The first output data D321 includes the interpretation result D3131 of the content indicated by the input data D101 by the first generating AI 510. Furthermore, the second output data D314 shown in Figure 12(b) is an example of the output data of the second generating AI 520, which takes input data D101, the second prompt D212 shown in Figure 6(b), and the first output data D313 shown in Figure 12(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D314 includes the interpretation result D3141 of the content indicated by input data D101 by the second generating AI 520, and the evaluation result D3142 of the validity of the interpretation result D3131 of the content indicated by input data D101 by the first generating AI 510. As can be seen by referring to evaluation result D3142 included in the second output data D314, the second generating AI 520 points out an error in the interpretation result of the first generating AI 510 regarding the content indicated by the input data D101. Furthermore, as can be seen by comparing the input data D101, the interpretation result D3131 included in the first output data D313, and the interpretation result D3141 included in the second output data D314, the accuracy of the interpretation of the content indicated by the input data D101 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0044] Figure 13 shows an example of the output data for the first generation AI510 and the second generation AI520 when the input data D102 shown in Figure 8 is applied. Specifically, the first output data D323 shown in Figure 13(a) is an example of the output data of the first generating AI 510, which takes input data D102 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D323 includes the interpretation result D3231 of the content indicated by the input data D102 by the first generating AI 510. Furthermore, the second output data D324 shown in Figure 13(b) is an example of output data from the second generating AI 520, which takes input data D102, the second prompt D212 shown in Figure 6(b), and the first output data D323 shown in Figure 13(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D324 includes the interpretation result D3241 of the content indicated by the input data D102 by the second generating AI 520, and the evaluation result D3242 of the validity of the interpretation result D3231 of the content indicated by the input data D102 by the first generating AI 510. As can be seen by referring to evaluation result D3242 included in the second output data D324, the second generating AI 520 points out an error in the interpretation result of the first generating AI 510 regarding the content indicated by the input data D102. Furthermore, as can be seen by comparing the input data D102, the interpretation result D3231 included in the first output data D323, and the interpretation result D3241 included in the second output data D324, the accuracy of the interpretation of the content indicated by the input data D102 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0045] Figure 14 shows examples of the output data for the first generation AI510 and the second generation AI520 when the input data D103 shown in Figure 10 is applied. Specifically, the first output data D333 shown in Figure 14(a) is an example of the output data of the first generating AI 510, which takes input data D103 and the first prompt D211 shown in Figure 6(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D333 includes the interpretation result D3331 of the content indicated by the input data D102 by the first generating AI 510. Furthermore, the second output data D334 shown in Figure 14(b) is an example of output data from the second generating AI 520, which takes input data D103, the second prompt D212 shown in Figure 6(b), and the first output data D333 shown in Figure 14(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D334 includes the interpretation result D3341 of the content indicated by input data D103 by the second generating AI 520, and the evaluation result D3342 of the validity of the interpretation result D3331 of the content indicated by input data D103 by the first generating AI 510. As can be seen by referring to evaluation result D3342 included in the second output data D334, the second generating AI 520 points out an error in the interpretation result of the first generating AI 510 regarding the content indicated by the input data D103. Furthermore, as can be seen by comparing the input data D103, the interpretation result D3331 included in the first output data D333, and the interpretation result D3341 included in the second output data D334, the accuracy of the interpretation of the content indicated by the input data D103 is improved when the first generating AI 510 and the second generating AI 520 are combined compared to when the first generating AI 510 is used alone.

[0046] The above describes Embodiment 2 of the information processing device according to one embodiment of the present disclosure with reference to Figures 12 to 14. As described above in this embodiment, even when a different combination of the first generation AI 510 and the second generation AI 520 is applied than in Embodiment 1 described above, it can be seen that an effect of improving the accuracy of interpretation of the content of input data can be expected compared to when the generation AI is used alone.

[0047] (Example 3) Referring to Figures 15 to 18, Embodiment 3 of an information processing device according to one embodiment of the present disclosure will be described. In this embodiment, an example will be described in which a different instruction from the second instruction D2122 of the second prompt D212 shown in Figure 6(b) is applied as the second instruction related to the evaluation of the validity of the interpretation result of the input data by the first generating AI, which is specified in the second prompt D202 shown in Figure 1. The combination of the first generating AI 510 and the second generating AI 520 will be the same as in Embodiment 1. In addition, the input data will be the same as in Embodiment 1, with input data D101 shown in Figure 5, input data D102 shown in Figure 8, and input data D103 shown in Figure 10 being applied.

[0048] Figure 15 shows an example of a prompt applied in this modified example. Specifically, the first prompt D221 shown in Figure 15(a) corresponds to an example of the first prompt D201 in the example shown in Figure 1, and includes a first instruction D2211 relating to the interpretation of the content indicated by the input data. The first prompt D221 is substantially the same as the first prompt D211 shown in Figure 6(a). Furthermore, the second prompt D222 shown in Figure 15(b) corresponds to an example of the second prompt D202 in the example shown in Figure 1, and includes a first instruction D2221 relating to the interpretation of the content indicated by the input data, and a second instruction D2222 relating to the evaluation of the validity of the interpretation result of the content indicated by the input data by the preceding generation AI (first generation AI 510). Note that, similar to the second prompt D212 shown in Figure 6(b), the first instruction D2221 included in the second prompt D222 is substantially the same as the first instruction D2211 included in the first prompt D221. On the other hand, in the second prompt D222, the instruction D2222 relating to the evaluation of the validity of the interpretation result of the content indicated by the input data by the preceding generation AI is expressed in a different way than the second instruction D2122 of the second prompt D212 shown in Figure 6(b).

[0049] Figure 16 shows an example of the output data for the first generation AI510 and the second generation AI520 when the input data D101 shown in Figure 5 is applied. Specifically, the first output data D315 shown in Figure 16(a) is an example of output data from the first generating AI 510, which takes input data D101 and the first prompt D221 shown in Figure 15(a) as inputs, and corresponds to the example of the first output data D301 shown in Figure 1. The first output data D315 includes the interpretation result D3151 of the content indicated by the input data D101 by the first generating AI 510, similar to the first output data D311 shown in Figure 7(a). Furthermore, the second output data D316 shown in Figure 16(b) is an example of output data from the second generating AI 520, which takes input data D101, the second prompt D222 shown in Figure 15(b), and the first output data D315 shown in Figure 16(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D316 includes the interpretation result D3161 of the content indicated by input data D101 by the second generating AI 520, and the validity evaluation result D3162 of the interpretation result D3151 of the content indicated by input data D101 by the first generating AI 510. As can be seen by comparing the interpretation result D3161 included in the second output data D316 shown in Figure 16(b) with the interpretation result D3121 included in the second output data D312 shown in Figure 7(b), in both cases it is possible to obtain an evaluation result of the validity of the interpretation result of the content indicated by the input data D101 by the first generating AI 510. Furthermore, similar to Example 1, the interpretation result D3161 included in the second output data D316 shows improved accuracy in interpreting the content of the input data D101 compared to the interpretation result D3151 included in the first output data D315, suggesting that the evaluation result D3162 has been taken into account in the interpretation result D3161.

[0050] Figure 17 shows an example of the output data for the first generation AI510 and the second generation AI520 when the input data D102 shown in Figure 8 is applied. Specifically, the first output data D325 shown in Figure 17(a) is an example of output data from the first generating AI 510, which takes input data D102 and the first prompt D221 shown in Figure 15(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D325 includes the interpretation result D3251 of the content indicated by input data D102 by the first generating AI 510, similar to the first output data D321 shown in Figure 9(a). Furthermore, the second output data D326 shown in Figure 17(b) is an example of output data from the second generating AI 520, which takes input data D102, the second prompt D222 shown in Figure 15(b), and the first output data D325 shown in Figure 17(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D326 includes the interpretation result D3261 of the content indicated by input data D102 by the second generating AI 520, and the evaluation result D3262 of the validity of the interpretation result D3251 of the content indicated by input data D102 by the first generating AI 510. As can be seen by comparing the interpretation result D3261 included in the second output data D326 shown in Figure 17(b) with the interpretation result D3221 included in the second output data D322 shown in Figure 9(b), in both cases it is possible to obtain an evaluation result of the validity of the interpretation result of the content indicated by the input data D102 by the first generating AI 510. Furthermore, similar to Example 1, the interpretation result D3261 included in the second output data D326 shows improved accuracy in interpreting the content of the input data D102 compared to the interpretation result D3251 included in the first output data D325, suggesting that the evaluation result D3262 has been taken into account in the interpretation result D3261.

[0051] Figure 18 shows an example of the output data for the first generation AI510 and the second generation AI520 when the input data D103 shown in Figure 10 is applied. Specifically, the first output data D335 shown in Figure 18(a) is an example of output data from the first generating AI 510, which takes input data D103 and the first prompt D221 shown in Figure 15(a) as inputs, and corresponds to an example of the first output data D301 shown in Figure 1. The first output data D335 includes the interpretation result D3351 of the content indicated by input data D103 by the first generating AI 510, similar to the first output data D331 shown in Figure 11(a). Furthermore, the second output data D336 shown in Figure 18(b) is an example of output data from the second generating AI 520, which takes input data D103, the second prompt D222 shown in Figure 15(b), and the first output data D335 shown in Figure 18(a) as inputs, and corresponds to an example of the second output data D302 shown in Figure 1. The second output data D336 includes the interpretation result D3361 of the content indicated by input data D103 by the second generating AI 520, and the validity evaluation result D3362 of the interpretation result D3351 of the content indicated by input data D103 by the first generating AI 510. As can be seen by comparing the interpretation result D3361 included in the second output data D336 shown in Figure 18(b) with the interpretation result D3321 included in the second output data D332 shown in Figure 11(b), in both cases it is possible to obtain an evaluation result of the validity of the interpretation result of the content indicated by the input data D103 by the first generating AI 510. Furthermore, similar to Example 1, the interpretation result D3361 included in the second output data D336 shows improved accuracy in interpreting the content of the input data D103 compared to the interpretation result D3351 included in the first output data D335, suggesting that the evaluation result D3362 has been taken into account in the interpretation result D3361.

[0052] The third embodiment of the information processing device according to one embodiment of the present disclosure has been described above with reference to Figures 15 to 18. As described above in this embodiment, the second instruction related to the evaluation of the validity of the interpretation result of the input data by the first generating AI, which is included in the second prompt, can be expressed in the same way as in the other embodiments, as long as the actual content remains the same.

[0053] (Example 4) Referring to Figures 19 and 20, Embodiment 4 of an information processing device according to one embodiment of the present disclosure will be described. In this embodiment, the results of a comparison of the accuracy of interpreting the content of input data will be summarized when the generation AI applied as the second generation AI on the downstream side is used alone and when the said generation AI is used in conjunction with another generation AI applied as the first generation AI on the upstream side. In this embodiment, the comparison will be between using Gemini provided by Google LLC alone and applying GPT-4o provided by OpenAI, Inc. as the first generation AI on the upstream side and applying Gemini as the second generation AI on the downstream side.

[0054] Figure 19 shows an example of the output data used for comparison in this embodiment. The output data D327 shown in Figure 19(a) is an example of output data when the target generation AI (Gemini) is used alone. Specifically, output data D327 corresponds to an example of output data of the target generation AI when the input data D102 shown in Figure 8 is applied as input data and the first prompt D211 shown in Figure 6(a) is applied as prompt. Output data D327 includes the interpretation result D3271 of the content indicated by the input data D102 by the target generation AI. Furthermore, the output data D328 shown in Figure 19(b) shows an example of output data when the target generation AI (Gemini) is applied as the second generation AI and another generation AI (GPT-4o) is applied as the first generation AI in the configuration shown in Figure 1. Specifically, output data D328 corresponds to an example of second output data (output data of the second generation AI) when input data D102 shown in Figure 8 is applied as input data, and the first prompt D211 and the second prompt D212 shown in Figure 6 are applied as the first and second prompts, respectively. Therefore, output data D328 includes the interpretation result D3281 of the content indicated by input data D102 by the second generation AI (Gemini) and the evaluation result D3282 of the validity of the interpretation result of the content indicated by input data D102 by the first generation AI (GPT-4o).

[0055] Here, we summarize the results of comparing the interpretation result D3271 contained in the output data D327 shown in Figure 19(a) and the interpretation result D3281 contained in the output data D328 shown in Figure 19(b) with the input data D102 shown in Figure 8. For example, in output data D327, the interpretation result for the answer to the question "On average, how much exercise do you do per week?" is "2," while the answer to the same question in input data D102 is "1," resulting in an error in the interpretation result. In contrast, in output data D328, the interpretation result for the answer to the same question is "1," which matches the answer to the same question in input data D102. Furthermore, output data D327 and output data D328 show consistent interpretations of the answers to questions other than the one mentioned above. From this, it can be seen that in this embodiment, when input data D102 is used as the target of processing, the accuracy of interpreting the content indicated by the input data D102 is improved when the generating AI (Gemini) is used in conjunction with other generating AIs compared to when the generating AI (Gemini) is used alone.

[0056] Figure 20 shows an example of the output data used for comparison in this embodiment. The output data D337 shown in Figure 20(a) is an example of output data when the target generation AI (Gemini) is used alone. Specifically, output data D337 corresponds to an example of output data of the target generation AI when the input data D103 shown in Figure 10 is applied as input data and the first prompt D211 shown in Figure 6(a) is applied as prompt. Output data D337 includes the interpretation result D3371 of the content indicated by the input data D103 by the target generation AI. Furthermore, the output data D338 shown in Figure 20(b) is an example of output data when the target generation AI (Gemini) is applied as the second generation AI and another generation AI (GPT-4o) is applied as the first generation AI in the configuration shown in Figure 1. Specifically, output data D338 corresponds to an example of second output data (output data of the second generation AI) when input data D103 shown in Figure 10 is applied as input data, and the first prompt D211 and the second prompt D212 shown in Figure 6 are applied as the first and second prompts. Therefore, output data D338 includes the interpretation result D3381 of the content indicated by input data D103 by the second generation AI (Gemini) and the evaluation result D3382 of the validity of the interpretation result of the content indicated by input data D103 by the first generation AI (GPT-4o).

[0057] Here, we summarize the results of comparing the interpretation result D3371 contained in the output data D337 shown in Figure 20(a) and the interpretation result D3381 contained in the output data D338 shown in Figure 20(b) with the input data D103 shown in Figure 10. For example, in output data D338, the interpretation result for the answer to the question "What is your residential area?" is "null," while the answer to the same question in input data D102 is "6," resulting in an error in the interpretation result. In contrast, in output data D338, the interpretation result for the answer to the same question is "6," which matches the answer to the same question in input data D103. As another example, in output data D338, the interpretation of the answer to the question, "Please tell us your satisfaction level with your current life and the reason why," is "8 / 10 Work is going well," whereas the answer to the same question in input data D102 is "8 / 10. Work is going well." Therefore, there is an error in the interpretation. In contrast, in output data D338, the interpretation of the answer to the same question is "8 / 10. Work is going well," which matches the answer to the same question in input data D103. Furthermore, output data D337 and output data D338 show consistent interpretations of the answers to questions other than the one mentioned above. From this, it can be seen that in this embodiment, when input data D103 is used as the target of processing, the accuracy of interpreting the content indicated by the input data D103 is improved when the target generation AI (Gemini) is used in conjunction with other generation AIs compared to when the target generation AI (Gemini) is used alone.

[0058] The third embodiment of the information processing apparatus according to one embodiment of the present disclosure has been described above with reference to Figures 19 and 20. As described above, in this embodiment as well as in the other embodiments, it can be expected that applying the information processing apparatus according to this embodiment will improve the accuracy of interpreting the content of the input data compared to using the generated AI alone.

[0059] As a supplement, it goes without saying that the effect of improving the accuracy of interpreting the content indicated by the input data may vary depending on the progress of learning of the first and second generative AIs applied (especially the progress of learning of the language data included as text information in the input data), as well as the ambiguity of the information contained in the input data.

[0060] For example, Figure 21 shows an example of output data when there is no difference in the interpretation of the content indicated by the input data between using the target generation AI alone and using that generation AI in conjunction with other generation AIs. The output data D317 shown in Figure 21(a) is an example of output data when the target generation AI (Gemini) is used alone. Specifically, output data D317 corresponds to an example of output data of the target generation AI when the input data D101 shown in Figure 5 is applied as input data and the first prompt D211 shown in Figure 6(a) is applied as prompt. Output data D317 includes the interpretation result D3171 of the content indicated by the input data D101 by the target generation AI. Furthermore, the output data D318 shown in Figure 21(b) is an example of output data when the target generation AI (Gemini) is applied as the second generation AI and another generation AI (GPT-4o) is applied as the first generation AI in the configuration shown in Figure 1. Specifically, output data D318 corresponds to an example of second output data (output data of the second generation AI) when input data D101 shown in Figure 5 is applied as input data, and the first prompt D211 and the second prompt D212 shown in Figure 6 are applied as the first and second prompts, respectively. Therefore, output data D318 includes the interpretation result D3181 of the content indicated by input data D101 by the second generation AI (Gemini) and the evaluation result D3182 of the validity of the interpretation result of the content indicated by input data D101 by the first generation AI (GPT-4o). In the example shown in Figure 21, as can be seen by comparing the interpretation result D3171 included in the output data D317 shown in Figure 21(a) with the interpretation result D3181 included in the output data D318 shown in Figure 21(b), there is no difference in the interpretation results of the input data. The above applies not only to this embodiment but also to other embodiments.

[0061] <Conclusion> As described above, an information processing device according to one embodiment of the present disclosure provides a first generating AI with input data including character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data, thereby causing the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data. The information processing device also provides a second generating AI with the above input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generating AI, thereby causing the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI, which is included in the first output data, and to output second output data including the result of interpreting the content of the character information contained in the input data, which reflects the evaluation result. With the above configuration, it is possible to obtain interpretation results with higher accuracy compared to when a single generating AI, such as the first or second generating AI, is used to interpret the content of the input data.

[0062] The embodiments described above are merely examples and do not necessarily limit the configuration or processing of the present invention. Various modifications and changes may be made without departing from the technical concept of the present invention. Furthermore, the scope of application of the embodiments described above is not particularly limited. That is, as long as the input data includes at least textual information such as a document, and the generating AI is made to interpret the content of the input data (in particular, the content of the textual information in the input data) from a specified perspective, the content of such interpretation is not particularly limited. As a specific example, the technology relating to this disclosure can also be applied to use cases in which data including a long text is used as input data, and the generating AI is made to interpret the content of the long text and create a summary. Furthermore, the type of input data is not particularly limited, as long as it contains textual information such as documents in a manner that the generating AI can recognize. Specifically, as mentioned above, image data containing textual information as an image may be used as input data, or data containing textual information as so-called text data may be used as input data. Furthermore, the combination of the first and second generating AIs is not particularly limited. For example, in Example 1, Gemini was applied as the first generating AI and Claude (a different type of generating AI than the first generating AI) was applied as the second generating AI, but a generating AI of the same type as the first generating AI may also be applied as the second generating AI. Also, even when the same type of generating AI is applied as the first and second generating AIs, generating AIs with different learning progresses may be applied, or substantially identical generating AIs generated by replication or the like may be applied. Furthermore, the present invention includes a program that realizes the functions of the embodiments described above, and a recording medium that can be read by a computer storing the program.

[0063] Furthermore, the following configurations also fall within the technical scope of this disclosure. (1) An information processing device comprising: a first interpretation processing unit that, upon inputting input data including character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data, causes the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data; and a second interpretation processing unit that, upon inputting the input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generating AI, causes the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI included in the first output data, and outputs second output data including the result of interpreting the content of the character information contained in the input data reflecting the evaluation result. (2) The information processing apparatus according to (1), wherein the second interpretation processing unit causes the second generation AI to output an evaluation result of the validity of the interpretation result of the content of the character information contained in the input data by the first generation AI. (3) The information processing device according to (1) or (2), wherein a different type of generation AI from the first generation AI is applied as the second generation AI. (4) The information processing apparatus according to any one of paragraphs (1) to (3), wherein the input data is data in which at least some character information is represented as an image, and the first instruction includes at least an instruction relating to the interpretation of the content of the character information represented as an image. (5) The information processing device according to (4), wherein the input data includes data that includes an image generated based on the results of scanning a document by a scanner. (6) Information processing method comprising: a first interpretation processing step in which a computer inputs input data including character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data to a first generating AI, thereby causing the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data; and a second interpretation processing step in which a computer inputs the input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generating AI, thereby causing the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI included in the first output data, and outputs second output data including the result of interpreting the content of the character information contained in the input data reflecting the evaluation result. (7) A program that causes a computer to perform a first interpretation processing step in which it inputs input data containing character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data, causing the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data; and a second interpretation processing step in which it inputs the input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generating AI, causing the second generating AI to evaluate the validity of the result of the interpretation of the content of the character information contained in the input data by the first generating AI included in the first output data, and outputs second output data including the result of interpreting the content of the character information contained in the input data that reflects the evaluation result. [Explanation of Symbols]

[0064] 1. Information Processing System 100 Information Processing Devices 101 Input Reception Section 102 Prompt generation unit 103 First Interpretation Processing Unit 104 Second Interpretation Processing Unit 510 The first generation AI 520 Second Generative AI

Claims

1. A first interpretation processing unit that, by inputting input data including character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data to a first generating AI, causes the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data. A second interpretation processing unit provides the second generation AI with input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generation AI, thereby causing the second generation AI to evaluate the validity of the interpretation result of the content of the character information contained in the input data by the first generation AI included in the first output data, and to output second output data including the interpretation result of the content of the character information contained in the input data that reflects the evaluation result. An information processing device having

2. The information processing apparatus according to claim 1, wherein the second interpretation processing unit causes the second generation AI to output an evaluation result of the validity of the interpretation result of the content of the character information contained in the input data by the first generation AI.

3. The information processing apparatus according to claim 1, wherein a generation AI of a different type from the first generation AI is applied as the second generation AI.

4. The aforementioned input data is data in which at least some character information is represented as an image. The first instruction includes at least instructions relating to the interpretation of the content of the textual information represented as the image. The information processing apparatus according to claim 1.

5. The information processing apparatus according to claim 4, wherein the input data includes data that includes an image generated based on the results of scanning a document by a scanner.

6. Computers A first interpretation processing step involves inputting input data containing character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data to a first generating AI, thereby causing the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data. A second interpretation processing step involves inputting the input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generation AI to a second generation AI, thereby causing the second generation AI to evaluate the validity of the interpretation result of the content of the character information contained in the input data by the first generation AI included in the first output data, and to output second output data including the interpretation result of the content of the character information contained in the input data that reflects the evaluation result. Information processing methods, including those mentioned above.

7. On the computer, A first interpretation processing step involves inputting input data containing character information and a first prompt including at least a first instruction relating to the interpretation of the content of the input data to a first generating AI, thereby causing the first generating AI to output first output data including the result of interpreting the content of the character information contained in the input data. A second interpretation processing step involves inputting the input data, the first output data, and a second prompt including at least the first instruction and a second instruction relating to the evaluation of the output of the first generation AI to a second generation AI, thereby causing the second generation AI to evaluate the validity of the interpretation result of the content of the character information contained in the input data by the first generation AI included in the first output data, and to output second output data including the interpretation result of the content of the character information contained in the input data that reflects the evaluation result. A program that executes something.

Images