Information processing device

Abstract

To provide an information processing device for realizing a question-and-answer process using an LLM in a manner that continuously poses questions in response to answers of a user.SOLUTION: The information processing device comprises: a sentence acceptance unit for accepting sentences input by a user; a prompt generation unit for, based on the sentence accepted by the sentence acceptance unit, generating a prompt to output a question sentence related to the sentence to a large language model; a prompt input unit for inputting the prompt generated by the prompt generation unit to a large language model; and a question output unit for outputting the question sentence output from the large language model as an answer to the prompt to the user as an answer to the sentence. The sentence acceptance unit further accepts a sentence input from the user as an answer to the question sentence output to the user by the question output unit.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] This specification discloses a technique related to an information processing device. [Background technology]

[0002] In recent years, information processing devices that utilize large-scale language models (LLMs) have become known, in which a prompt, which is a question in natural language, is input into the LLM and a response in natural language is obtained from the LLM (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7441366 Summary of the Invention [Problem to be solved by the invention]

[0004] There has been insufficient consideration given to the format of question and answer sessions using LLM. [Means for solving the problem]

[0005] The technology disclosed in this specification can be realized in the following forms.

[0006] (1) One aspect disclosed in this specification is an information processing device. The information processing device includes: a sentence accepting unit that accepts a sentence input by a user; a prompt generating unit that generates a prompt to cause a large-scale language model to output a question related to the sentence based on the sentence accepted by the sentence accepting unit; a prompt input unit that inputs the prompt generated by the prompt generating unit to the large-scale language model; and a question output unit that outputs the question output from the large-scale language model in response to the prompt to the user as a response to the sentence. The sentence accepting unit further accepts a sentence input by the user as a response to the question output to the user by the question output unit. This aspect of the information processing device enables a question-and-answer session using an LLM in which questions are continuously asked in response to answers from the user.

[0007] (2) In the information processing device of the above aspect, the prompt may include information specifying at least one of the attributes of the large-scale language model as a questioner when outputting the question and the question format of the question. With this information processing device, the question output from the LLM can be adjusted to the desired format.

[0008] (3) In the information processing device of the above aspect, the prompt may include an instruction to cause the large-scale language model to output a question regarding an error in the sentence if the sentence contains an error. This information processing device can reduce the number of inappropriate questions output from the LLM due to an error in the sentence entered by the user.

[0009] The technology disclosed in this specification can be realized in various forms other than an information processing device, such as an information processing method or a computer program. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram illustrating a configuration of a chat system. [Figure 2] FIG. 2 is an explanatory diagram showing the physical configuration of a user terminal. [Figure 3] FIG. 2 is an explanatory diagram showing the functional configuration of a user terminal. [Figure 4] 10 is a flowchart showing chat processing. [Figure 5] FIG. 10 is an explanatory diagram showing an example of a chat. [Figure 6] FIG. 10 is an explanatory diagram illustrating an example of a prompt. [Figure 7] FIG. 10 is an explanatory diagram showing an example of a chat. DETAILED DESCRIPTION OF THE INVENTION

[0011] 1 is an explanatory diagram showing the configuration of a chat system 10. The chat system 10 is a system that realizes chat with users by utilizing large language models (LLMs). In the chat system 10, as a response to a sentence input by a user, a question sentence related to the sentence is output to the user.

[0012] The chat system 10 includes a user terminal 100, a computer network 300, and an LLM system 500. The user terminal 100 is an information processing device configured to be operable by a user of the chat system 10. The user terminal 100 may be a smartphone, a personal computer, or a tablet terminal.

[0013] The computer network 300 communicably connects the user terminal 100 and the LLM system 500. The computer network 300 includes at least one of a mobile communication network, the Internet, a wireless LAN, and a wired LAN.

[0014] The LLM system 500 is a computer system equipped with an LLM model. The LLM system 500 receives prompts from the user terminal 100 via the computer network 300, and provides the user terminal 100 with sentences output from the LLM based on the prompts.

[0015] 2 is an explanatory diagram showing the physical configuration of the user terminal 100. The user terminal 100 includes a processor 110, a memory 120, a user interface 130, and a communication interface 140.

[0016] The processor 110 of the user terminal 100 performs various processes by executing a plurality of program instructions stored in the memory 120. The number of processors 110 may be one or more.

[0017] The memory 120 of the user terminal 100 stores data handled by the processor 110. The number of memories 120 may be one or more. The memory 120 stores a chat application 124 and configuration data 126. The chat application 124 is a computer program for realizing chats with users using the LLM. The configuration data 126 includes various configuration information used to execute the chat application 124 (e.g., basic sentences for prompts, information to be selectively added to prompts, etc.).

[0018] The user interface 130 of the user terminal 100 is configured to be able to accept input of text information from the user. The user interface 130 is configured to be able to output text information to the user. In this embodiment, the user interface 130 is a touch panel display configured to be able to accept input of text information from the user and to be able to display text information to the user. The user interface 130 may be configured by appropriately combining a touch panel display, a keyboard, a display, a microphone, a speaker, and the like.

[0019] The user interface 130 is configured to be able to display a chat screen CS. The chat screen CS is a GUI (Graphical User Interface) for displaying chats with users who use the LLM. The chat screen CS is configured to be able to receive text ST from a user. The chat screen CS is configured to be able to display a question QT output from the LLM system 500 as a response to the text ST.

[0020] The communication interface 140 of the user terminal 100 is configured to be able to communicate with the LLM system 500 via the computer network 300. In this embodiment, the communication interface 140 includes a wireless circuit connectable to a mobile phone network and a wireless circuit compliant with the wireless LAN standard.

[0021] 3 is an explanatory diagram showing the functional configuration of the user terminal 100. The user terminal 100 includes a text accepting unit 182, a prompt generating unit 184, a prompt input unit 186, a question receiving unit 192, and a question output unit 196.

[0022] The text acceptance unit 182 of the user terminal 100 is a functional unit that accepts text ST input by a user. The text acceptance unit 182 is realized by the processor 110 executing program instructions included in the chat application 124. By making the text acceptance unit 182 function, the user terminal 100 accepts text ST input by a user via a chat screen CS displayed on the user interface 130.

[0023] The prompt generation unit 184 of the user terminal 100 is a functional unit that generates a prompt PT based on a sentence ST accepted by the sentence acceptance unit 182, causing the LLM system 500 to output a question sentence QT related to the sentence ST. The prompt generation unit 184 is realized by the processor 110 executing program instructions included in the chat application 124. The user terminal 100 causes the prompt generation unit 184 to function, thereby generating a prompt PT based on the sentence ST.

[0024] The prompt PT may include information specifying at least one of the attributes of the questioner that the LLM system 500 will act as when outputting the question QT and the question format of the question QT. The attributes of the questioner included in the prompt PT may be occupation, age, gender, personality, nationality, etc. The question format included in the prompt PT may be open questions, closed questions, scaling questions, coping questions, miracle questions, etc. The prompt PT may include an instruction to cause the LLM system 500 to output a question regarding an error in the sentence ST if the sentence ST contains an error.

[0025] The prompt input unit 186 of the user terminal 100 is a functional unit that inputs the prompt PT generated by the prompt generation unit 184 to the LLM system 500. The prompt input unit 186 is realized by the processor 110 executing program instructions included in the chat application 124. By operating the prompt input unit 186, the user terminal 100 inputs the prompt PT to the LLM system 500 via the computer network 300 using the communication interface 140.

[0026] The question receiving unit 192 of the user terminal 100 is a functional unit that receives a question QT output from the LLM system 500 as a response to a prompt PT. The question receiving unit 192 is realized by the processor 110 executing program instructions included in the chat application 124. By operating the question receiving unit 192, the user terminal 100 receives the question QT from the LLM system 500 via the computer network 300 using the communication interface 140.

[0027] The question output unit 196 of the user terminal 100 is a functional unit that outputs a question QT to the user as a response to the text ST. The question output unit 196 is realized by the processor 110 executing program instructions included in the chat application 124. By causing the question output unit 196 to function, the user terminal 100 outputs the question QT to the user as a response to the text ST via the chat screen CS displayed on the user interface 130.

[0028] After functioning as the question output unit 196, the user terminal 100 again functions as the text receiving unit 182. By causing the text receiving unit 182 to function, the user terminal 100 further receives new text ST input by the user as a response to the question text QT output to the user by the question output unit 196 via the chat screen CS displayed on the user interface 130.

[0029] Fig. 4 is a flowchart showing chat processing. The chat processing in Fig. 4 is processing for realizing a chat with a user using LLM. When the processor 110 executes program instructions of the chat application 124, the user terminal 100 starts the chat processing in Fig. 4.

[0030] After starting the chat process of FIG. 4, the user terminal 100 waits until the user inputs text ST on the chat screen CS of the user interface 130 (step S110, step S115: "NO").

[0031] If the user inputs a sentence ST on the chat screen CS (step S115: "YES"), the user terminal 100 accepts the sentence ST input by the user on the chat screen CS (step S120). After that, the user terminal 100 generates a prompt PT that causes the LLM system 500 to output a question sentence QT regarding the sentence ST input by the user (step S130).

[0032] After generating the prompt PT (step S130), the user terminal 100 inputs the prompt PT to the LLM system 500 via the computer network 300 using the communication interface 140 (step S140). Thereafter, the user terminal 100 acquires the question QT output from the LLM system 500 as a response to the prompt PT (step S150).

[0033] After acquiring the question QT (step S150), the user terminal 100 displays the question QT on the chat screen CS of the user interface 130 (step S160). Thereafter, the user terminal 100 waits again until the user inputs a sentence ST on the chat screen CS of the user interface 130 (steps S110, S115: "NO"). In this way, the display of the question QT and the input of the sentence ST are repeated, thereby realizing a chat with the user using the LLM on the chat screen CS.

[0034] Figure 5 is an explanatory diagram showing an example of a chat. Figures 5(a), 5(b), and 5(c) respectively show a sentence ST input by a user, a prompt PT based on the sentence ST, and a question QT based on the prompt PT.

[0035] Sentence ST in Figure 5(b) is a user's response to question QT in Figure 5(a). Sentence ST in Figure 5(c) is a user's response to question QT in Figure 5(b). In this way, by repeatedly displaying question QT and inputting sentence ST, a chat with a user using LLM is realized.

[0036] The prompt PT in Figure 5 includes an instruction to cause the LLM system 500 to output a question regarding an error if there is an error in the sentence ST. Specifically, the prompt PT in Figure 5 includes the instruction, "If there is an obvious error in the content below, please ask about the error again."

[0037] Fig. 6 is an explanatory diagram showing an example of a prompt PT. Fig. 6(a) to Fig. 6(e) show examples of prompt PTs with different patterns, each with different questioner attributes and question formats. The prompt pattern may be a specification that is fixedly set in the user terminal 100, a specification that can be selectively set by the user, or a specification that is automatically set according to the content of the text ST.

[0038] 7A, 7B, and 7C are explanatory diagrams showing an example of a chat. Each of the diagrams shows a sentence ST input by a user, a prompt PT based on the sentence ST, and a question QT based on the prompt PT.

[0039] Sentence ST in Figure 7(b) is a user's response to question QT in Figure 7(a). Sentence ST in Figure 7(c) is a user's response to question QT in Figure 7(b). In the example of Figure 7, the LLM system 500 is designed to be able to respond by taking into account the chat history, so the prompt PT in Figure 7(a) causes the LLM system 500 to output a question QT regarding the following sentence ST. In prompts PT from Figure 7(b) onwards, the sentence ST input by the user is used as the prompt as is, and the question QT is displayed as the response.

[0040] The user terminal 100 described above can realize a question and answer session using the LLM system 500 in a manner in which questions are continually asked in response to answers from the user. Furthermore, the user terminal 100 can adjust the question QT output from the LLM system 500 to an intended manner by including in the prompt PT information specifying at least one of the questioner's attributes and the question format of the question QT. Furthermore, the user terminal 100 can reduce the output of inappropriate questions from the LLM system 500 due to errors in the sentence ST by including in the prompt PT an instruction to output a question QT related to the error if there is an error in the sentence ST.

[0041] The technology disclosed in this specification is not limited to the above-described embodiments, examples, and modifications, and can be realized in various configurations without departing from the spirit thereof. For example, among the technical features of the above-described embodiments, examples, and modifications, those corresponding to the technical features of each form described in the Summary of the Invention section can be appropriately replaced and combined to solve some or all of the above-described problems or achieve some or all of the above-described effects. Furthermore, technical features not described as essential in this specification can be appropriately deleted.

[0042] For example, at least a part of the functions of the user terminal 100 may be realized as functions provided by another information processing device (server) connected to the computer network 300. [Explanation of symbols]

[0043] 10. Chat system 100...User terminal 110...Processor 120...Memory 124...chat application 126...Setting data 130...User Interface 140...Communication interface 182…Text Reception Section 184...Prompt generation unit 186...Prompt input section 192...Question Receiving Section 196...Question output section 300...Computer Networks 500...LLM system

Claims

1. An information processing device, a text receiving unit that receives text input by a user; a prompt generation unit that generates, based on the sentence accepted by the sentence acceptance unit, a prompt that causes a large-scale language model to output a question sentence related to the sentence; a prompt input unit that inputs the prompt generated by the prompt generation unit into the large-scale language model; a question output unit that outputs a question output from the large-scale language model as a response to the prompt to the user as a response to the sentence; Equipped with The text receiving unit further receives a text input by the user as a response to the question text output to the user by the question text output unit.

2. 2. The information processing device according to claim 1, The information processing device, wherein the prompt includes information specifying at least one of an attribute of the large-scale language model as a questioner when outputting the question sentence, and a question format of the question sentence.

3. 3. The information processing device according to claim 1, The information processing device, wherein the prompt includes an instruction to cause the large-scale language model to output a question regarding an error if the sentence contains an error.

Images