Apparatus, method, and program for a generative AI model to reflect unlearned information.

Abstract

When a generative AI model incorporates previously unlearned information into its response to user input, it enhances the relevance of that information to the input. [Solution] An application used by a user receives as relevant background information a portion of a set of elements that constitute a predetermined format of background information associated with the user or the group to which the user belongs, which is either input from the user or a portion identified according to the user making the input. The application then makes a request to a generating AI model that includes the relevant background information and provides the user with an output using the response to that request.

Description

Technical Field

[0001] The present invention relates to an apparatus, a method, and a program for enabling a generative AI model to reflect unlearned information.

Background Art

[0002] In order to improve the quality of the output of generative AI models such as large language models (LLMs) whose use is expanding, methods have been tried to make the generative AI model consider information that the generative AI model has not learned, such as in-house information of the user company. As an example, in a method called RAG (Retrieval-Augmented Generation), but not limited to this, each part called a chunk obtained by dividing each unlearned document is vectorized and stored in a vector database, and one or more chunks semantically close to the input from the user are retrieved. Then, by including those chunks obtained as search results in the input to the generative AI model, the unlearned information is reflected in the output from the generative AI model.

Summary of the Invention

Problems to be Solved by the Invention

[0003] However, when storing various documents and other data in a database such as a vector database, it becomes difficult to obtain data highly relevant to the input from the user as search results. For example, when storing in-house information, old documents may remain when the document is updated and there is a new version, and the latest version may not necessarily be given to the generative AI model. Generally speaking, there may be a situation where a plurality of mutually inconsistent pieces of information are stored. Also, as the number of documents to be stored increases, files with similar file names or files with similar contents also increase, and there may be a case where the file to be referred to is not referred to.

[0004] The present invention has been made in view of these points, and its objective is to improve the relevance between the input and the information given to the generating AI model within the unlearned information, in an apparatus, method, or program for which the output of a generating AI model in response to user input reflects information that the generating AI model has not yet learned. [Means for solving the problem]

[0005] To achieve this objective, a first aspect of the present invention provides a method for causing a generating AI model to reflect unlearned information in the output of the generating AI model in response to user input, the method comprising: an application obtaining the input from the user; the application transmitting a request for the input or relevant background information corresponding to the user making the input; the application receiving some or all of a plurality of elements having a predetermined form of background information associated with the user or a group to which the user belongs as relevant background information; the application making a request to the generating AI model including the input and the relevant background information; and the application providing the user with the output using the response to the request.

[0006] Furthermore, a second aspect of the present invention is the method of the first aspect, wherein each of the plurality of elements has one or more attributes.

[0007] Furthermore, a third aspect of the present invention is the method according to the first or second aspect, wherein each of the plurality of elements has an ID.

[0008] Furthermore, a fourth aspect of the present invention is a method according to any of the first to third aspects, wherein a first element included in the plurality of elements is linked to a second element included in the plurality of elements.

[0009] Furthermore, a fifth aspect of the present invention is a method according to any of the first to third aspects, wherein each of the plurality of elements is linked to other elements included in the plurality of elements.

[0010] Furthermore, a sixth aspect of the present invention is a method according to any of the first to fifth aspects, wherein the related background information includes a portion of the plurality of elements that have a relatively high relevance to the input.

[0011] Furthermore, a seventh aspect of the present invention is a method according to any of the first to fifth aspects, wherein the relevant background information includes a portion of the plurality of elements that are relatively more relevant to the user.

[0012] Furthermore, an eighth aspect of the present invention is the method of the seventh aspect, wherein the related background information includes a portion of the plurality of elements that have a relatively high relevance to past inputs by the user.

[0013] Furthermore, a ninth aspect of the present invention is a method according to any six to eighth aspect, wherein the related background information further comprises one or more elements having association with at least one of the aforementioned parts.

[0014] Furthermore, a tenth aspect of the present invention is a method according to any of the sixth to eighth aspects, wherein the related background information further comprises one or more elements having a link with at least one of the aforementioned parts, wherein the type of the link is a predetermined type.

[0015] Furthermore, an eleventh aspect of the present invention is a method according to any of the sixth to eighth aspects, wherein the related background information further includes one or more elements having a linkage with at least one of the aforementioned parts, the strength of which the linkage is greater than or equal to a predetermined value.

[0016] Furthermore, a twelfth aspect of the present invention is a method according to any one of the first to eleventh aspects, wherein the related background information includes metadata of resources related to the input.

[0017] Furthermore, a thirteenth aspect of the present invention is a method according to any one of the first to eleventh aspects, wherein the related background information includes metadata of a document related to the input.

[0018] Furthermore, a fourteenth aspect of the present invention is the method of the thirteenth aspect, wherein the metadata includes a storage location for documents related to the input.

[0019] Furthermore, a fifteenth aspect of the present invention is a method according to any of the first to eleventh aspects, wherein the related background information includes metadata of a person related to the input.

[0020] Furthermore, a sixteenth aspect of the present invention is the method of the fifteenth aspect, wherein the metadata includes the contact information of the person.

[0021] Furthermore, a 17th aspect of the present invention is a method according to any of the first to 11th aspects, wherein the related background information includes metadata of the AI ​​assistant related to the input.

[0022] Furthermore, an eighteenth aspect of the present invention is the method of the seventeenth aspect, wherein the metadata includes information for executing an AI assistant related to the input.

[0023] Furthermore, a 19th aspect of the present invention is a method according to any of the first to 18th aspects, wherein the predetermined format is a format using a predetermined data description language.

[0024] Furthermore, a 20th aspect of the present invention is a method according to any of the first to 19 aspects, wherein the requirement includes basic information associated with the user or the group.

[0025] Furthermore, a 21st aspect of the present invention is the method of the 20th aspect, wherein the basic information includes one or more terms specific to the business of the user or the group.

[0026] Moreover, a 22nd aspect of the present invention is the method according to any one of the 1st to 21st aspects, wherein the request includes basic information associated with the application.

[0027] Moreover, a 23rd aspect of the present invention is a program for causing a computer to execute a method for causing a generation AI model to reflect unlearned information in an output from the generation AI model according to an input from a user, the method including: a step in which an application acquires the input from the user; a step in which the application transmits a request for acquiring related background information according to the input or the user who makes the input; a step in which the application receives, as the related background information, some or all of a plurality of elements included in background information in a predetermined format associated with the user or a group to which the user belongs; a step in which the application makes a request including the input and the related background information to the generation AI model; and a step in which the application provides the user with the output using the response to the request.

[0028] Moreover, a 24th aspect of the present invention is an apparatus for causing a generation AI model to reflect unlearned information in an output according to an input from a user, the apparatus being configured to acquire the input from the user, transmit a request for acquiring related background information according to the input or the user who makes the input, receive, as the related background information, some or all of a plurality of elements included in background information in a predetermined format associated with the user or a group to which the user belongs, make a request including the input and the related background information to the generation AI model, and provide the user with the output using the response to the request.

[0029] ​​Furthermore, a 25th aspect of the present invention is a method for providing background information, comprising the steps of: receiving a request from an application to acquire relevant background information from pre-stored background information, which is input from a user to the application or related background information corresponding to a user making an input; creating some or all of a plurality of elements having background information of a predetermined format associated with an identifier of the user or group to which the user belongs, included in the acquisition request, as the relevant background information; and transmitting the relevant background information to the application.

[0030] Furthermore, a 26th aspect of the present invention is a method according to the 25th aspect, wherein the creation step includes the step of searching a first database for a portion of the plurality of elements that have a relatively high similarity to the input, and the step of searching a second database for one or more elements that have a link with at least one of the portion.

[0031] Furthermore, a 27th aspect of the present invention is the method of the 26th aspect, wherein the first database is a vector database.

[0032] Furthermore, a 28th aspect of the present invention is the method of the 26th aspect, wherein the step of searching for a portion includes the step of searching for a portion that has a relatively high similarity to one or more keywords included in the input.

[0033] Furthermore, a 29th aspect of the present invention is a method according to the 25th aspect, wherein the creation step includes a step of searching for background information associated with the identifier, and a step of extracting a portion of a plurality of elements having the background information, wherein a first element included in the portion is associated with a second element included in the portion.

[0034] Furthermore, a 30th aspect of the present invention is a method according to the 29th aspect, wherein the extraction step includes a step performed based on one or more keywords included in the input.

[0035] Furthermore, a 31st aspect of the present invention is a method according to any 25th to 30th aspect, further comprising the step of excluding at least one of a plurality of inconsistent elements from the relevant background information.

[0036] Furthermore, a 32nd aspect of the present invention is a program for causing a computer to perform a method for providing background information, the method comprising: receiving a request from an application to acquire relevant background information from pre-stored background information, which is input from a user to the application or information corresponding to a user performing the input; creating relevant background information by taking some or all of a plurality of elements having background information of a predetermined format associated with an identifier of the user or group to which the user belongs, which is included in the acquisition request; and transmitting the relevant background information to the application.

[0037] A 33rd aspect of the present invention is a device for providing background information, which receives input from a user to an application or a request to acquire relevant background information corresponding to the user making the input, from among pre-stored background information, creates a part or all of a plurality of elements of background information in a predetermined format associated with an identifier of the user or the group to which the user belongs included in the acquisition request as relevant background information, and transmits the relevant background information to the application. [Effects of the Invention]

[0038] According to one aspect of the present invention, an application used by a user receives as relevant background information a portion of a plurality of elements of background information in a predetermined format associated with the user or the group to which the user belongs, which is input from the user or a portion identified according to the user making the input, and makes a request to the generating AI model that includes the relevant background information, and provides the user with an output using the response to the request, thereby increasing the relevance of unlearned information provided to the generating AI model in generating output in response to user input by the generating AI model. [Brief explanation of the drawing]

[0039] [Figure 1] This is a diagram showing a system according to one embodiment of the present invention. [Figure 2] This diagram shows the flow of a method according to one embodiment of the present invention. [Figure 3] This figure shows an example of background information relating to one embodiment of the present invention. [Figure 4] This diagram shows the flow of a method according to one embodiment of the present invention from the perspective of the context server. [Modes for carrying out the invention]

[0040] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0041] Figure 1 shows a system according to one embodiment of the present invention. The device 100 provides information referred to herein as "context information" to an application server 120 that provides an application that generates output using a generation AI model in response to input from a user terminal 110 used by the user, via an IP network such as the Internet. In the example in Figure 1, the application server 120 uses the generation AI model by communicating with a platform 130 that provides the generation AI model via an IP network. However, an application for providing the generation AI model may be executed on the application server 120 so that the generation AI model is also provided by the application server 120, or the application provided by the application server 120 itself may provide the generation AI model. Hereinafter, in this specification, the device 100 may be referred to as a "context server" because it provides context information.

[0042] Although Figure 1 shows only a single application server 120, the device 100 can provide user background information to multiple application servers. User or group background information may be common to multiple applications, or it may be managed separately for each application. User terminals 110 may communicate with the device 100 over an IP network as needed to add, delete, or modify their own background information.

[0043] If high security is required, the device 100 may communicate with the user terminal 110 or application server 120 via a closed network such as an intranet or local area network.

[0044] The device 100 comprises a communication unit 101 such as a communication interface, a processing unit 102 such as a processor or CPU, and a storage unit 103 including a storage device or storage medium such as memory or a hard disk. The device can be configured by executing a program for performing each process or operation in the processing unit 102. The device 100 may include one or more devices, computers, or servers. The program may also include one or more programs and can be recorded on a computer-readable storage medium to form a non-transient program product. The program is stored in the storage unit 103 or in a storage device or storage medium such as a database 104 accessible via an IP network, and at least one processor in the processing unit 102 can execute the instructions contained in the program. The data described below as being stored in the storage unit 103 may be stored in a storage device or storage medium such as a database 104, and vice versa. Figure 1 shows the storage device 104 as an example, divided into a first storage device 104-1 and a second storage device 104-2; further details will be described later. Furthermore, the user terminal 110 and the application server 120 can also be configured to have a similar communication unit, processing unit, and storage unit.

[0045] First, the application server 120 receives user input from the user terminal 110 (S201). This input may be text data, audio data, or image data, or any combination thereof. Although this explanation assumes that the application server 120 provides the application, the application, or the application and the generative AI model used by the application, may also be provided on the user terminal 110. In that case, the user terminal 110 acquires the input by accepting user input for itself.

[0046] User input can be performed from an input screen displayed by sending input screen display information to the user terminal 110. The input screen display information can be sent, for example, as an HTML file, read by the user terminal 110's web browser, and displayed on the user terminal 110's display screen. If a dedicated application is installed on the user terminal 110, the input screen display information should include the data necessary for that application to display the input screen. Here, "input screen" can take various forms such as a web page, modal window, or popup window when displayed on a web browser, or a screen of a dedicated application when displayed on that application. In any case, any screen containing an area with input fields for entering necessary information qualifies as an input screen. Similar techniques can be applied to other screens mentioned in this specification.

[0047] Next, the context server 100 receives a request from the user terminal 110 or the application server 120 to acquire relevant background information corresponding to the input (S202). Since the context server 100 can manage background information for each user or each group to which that user belongs, subsequent processing becomes possible when the acquisition request includes an identifier for the user who made the input or the group to which that user belongs. The identifier included in the acquisition request and the identifier used by the context server 100 to manage background information for each user or group do not need to be exactly the same; it is sufficient that they are associated with each other, and in the following explanation, we will not distinguish between them even if they are different. Furthermore, the relevant background information may be specific to the user making the input, or more specifically, it may be specific to the user's past inputs. Such past inputs may be included in the acquisition request or stored in the context server 100.

[0048] A user or the group to which that user belongs must grant the application running on the user terminal 110 or application server 120 the right to access the user's background information managed by the context server 100. If the user is logged into the application when making input and the above access rights have been granted, the acquisition request may include the identifier of the user or the group to which that user belongs, which is used by the context server 100 to manage background information. If the context server 110 manages the background information of a user or the group to which that user belongs on an application-by-application basis, the acquisition request may further include the identifier of the application to which the user made input. The identifier of the user or the group to which that user belongs and the identifier of the application may not be separate identifiers, but rather a single identifier that includes all of this information.

[0049] The context server 100 creates related background information from a portion of the elements of one or more pieces of background information associated with the received identifier. Specifically, in one example, the context server 100 may identify a portion of the elements of the background information associated with the received identifier that have a relatively high relevance to the input from the storage device 104, which is a database (S203), and use that portion as related background information. This identification may be performed using a generative AI model.

[0050] Background information has a predetermined format, and more specifically, it can be in a format using a predetermined data description language such as JSON or YAML. Background information has multiple elements, and each element has one or more attributes. Each element may also have an ID. Furthermore, the elements of the background information may include the time of addition or modification, and may also include version information if it has been modified.

[0051] Figure 3 shows an example of background information according to one embodiment of the present invention. This background information is the user's background information referenced in an AI assistant application used by the user, and the element with id 0 describes several attributes of the AI ​​assistant. This element may be stored as basic information separate from the user's background information, as it is something that the AI ​​assistant should always refer to. Here, in this specification, "AI assistant" refers to a program that generates output according to input using an AI model. The element with id 1 describes that the user's name is "X," and further describes several attributes that the user possesses. The element with id 3 is about the concept of "Koopman operator," and its attributes describe that it is of interest to user "X," which is the element with id 1, and it is linked to the element with id 1. The element with id 2 is about a paper on "Koopman RL," and its storage location is described as an attribute. Furthermore, this element describes as discussing a system using the concept of the element with id 3, and it is linked to the element with id 3.

[0052] As shown in the example in Figure 3, background information can have metadata, such as the location where the document is stored, as an attribute of the element. In this way, by making the metadata, rather than the content of the document itself, an attribute of the background information element, the metadata should be included in the request to the generating AI model when there is a high need to refer to the document in relation to user input. In other words, although it is not necessary to exclude it completely, it is preferable that the background information does not include all or part of the content of the various documents themselves.

[0053] Furthermore, background information can include metadata about a person as an attribute of an element, such as the description of the element with id 1 as someone studying “sub-Gaussianity” at university. This metadata may also include the person's contact information. When it is appropriate or beneficial to contact a specific person in relation to user input, including metadata about that person in the request to the generating AI model allows the generating AI model to generate output that includes what to ask that person and their contact information.

[0054] More generally, if background information includes metadata for people, documents, or other resources, and relevant background information in response to input to an application includes metadata for resources related to that input, then such metadata may include information for accessing those resources.

[0055] Furthermore, a first element included in a plurality of elements of background information may be linked to a second element included in the plurality of elements, and each of the plurality of elements may also be linked to other elements included in the plurality of elements. Background information may also have elements linked to elements of other background information.

[0056] In another example of creating related background information, the context server 100 can first retrieve from the first storage device 104-1, which is a vector database, a portion of the background information associated with the received identifier that has a relatively high similarity to the input, and then retrieve from the second storage device 104-2, which is a relational database, one or more elements that have a link with at least one of the extracted portions, to create related background information. The links between elements may be divided into multiple types and may also be divided into multiple strengths. Specifically, the first storage device 104-1 and the second storage device 104-2 can be databases other than those described herein. It is preferable to divide the background information into parts such as element-specific chunks and vectorize each element. The search from the first storage device 104-1 can be a search of a portion that has a relatively high similarity to one or more keywords included in the input.

[0057] In yet another example, the context server 100 first retrieves background information associated with the received identifier from the storage device 104. Then, the context server 100 extracts some of the elements of the obtained background information. This extraction can be performed based on one or more keywords included in the user input. At least one of these elements has a link to other elements. The linking between elements may be divided into multiple types and also into multiple strengths. This extraction may be performed using a generative AI model.

[0058] Next, the context server 100 receives relevant background information from the storage device 104 (S204) and transmits this relevant background information to the user terminal 110 or the application running on the application server 120 (S205). Here, the same term "relevant background information" is used for both the data received and the data transmitted by the context server 100, but this is not intended to exclude the possibility that the context server 100 may perform processing such as data format conversion, addition, deletion, or modification of parts of the data.

[0059] Related background information is a part of the multiple elements of background information associated with a user or the group to which the user belongs, and at least one of these parts has a link with other elements. The other elements may have a predetermined type of link, and the strength of the link may be greater than or equal to a predetermined value. If there are multiple elements that do not match each other based on the link between the two, the context server 100 may exclude at least one of them from the related background information. Such exclusion processing may be performed on the user terminal 110 or application server 120 on which the application is executed. In particular, between elements of different background information, it may be necessary to exclude one of them. Such exclusion processing may be performed using a generative AI model. Although related background information has been described as a part of the multiple elements of background information, if the data size of the background information is not large, the entirety may be considered related background information.

[0060] The context server 100 may record the number of times each of the multiple elements included in the background information of a user or the group to which the user belongs has been included in the created related background information, i.e., the number of references. This number of references can be used for weighting in evaluating the relevance with user input.

[0061] When an application runs on the application server 120, the application server 120 makes a request to the generating AI model that includes at least a portion of the input from the user and the relevant background information received (S206), and receives a response to the request (S207). For example, by including at least a portion of the relevant background information corresponding to the input in the request, the generating AI model can dynamically reference the knowledge that the user has or should have, or that has been shared or should be shared within the group to which the user belongs, and generate an output that is useful for the input.

[0062] The request may include background information associated with the user or the group to which the user belongs, regardless of user input. This background information may include, for example, one or more terms specific to the work of the user or the group. The request may also include background information associated with the applications used by the user.

[0063] An application that receives a response from a generative AI model to user input may verify the response based on relevant background information. That is, if the relevant background information has many elements, the response from the generative AI model is expected to be based on at least a part of that relevant background information and therefore reflects it. However, if the response actually obtained is not, the application may modify the acquired relevant background information or acquire the relevant background information again and make a request to the generative AI model that includes at least a part of the new relevant background information.

[0064] In this specification, "AI model" means a machine learning model that has been trained to predict an output for a given input, and "generative AI model" means a large-scale language model (LLM) that has been trained using text data to generate an output not included in the input for a given input. As a generative AI model, an LLM applying a transformer architecture is particularly preferred, but it is expected that the name of the architecture may change as technology advances. Therefore, in this specification, "transformer architecture" includes architectures that use one or more features of the transformer architecture or improvements thereof. In this specification, whether or not "generative AI models" are the same is determined by whether or not the type of generative AI model specified by the user is the same. In the case of the Open AI API, for GPT-4, if the value of the variable "model" is the same, it is expressed as the same generative AI model. If generative AI models used in different processes are not the same, they may be provided on the same platform 130 or on different devices. Needless to say, a request for a generative AI model may include multiple requests, and may include the execution of one or more programs for one or more processes performed on the device making the request, other than calling the generative AI model. These programs can be understood as part of applications such as AI assistants provided on the device.

[0065] A generative AI model can be invoked by executing code containing instructions (prompts) written in natural language, thereby calling the OpenAI API. The generative AI model then generates a response to these prompts. The OpenAI API is an example, and other APIs may be used. More specifically, the code for making the call can be stored in the memory of the device on which the code is executed. The device can then retrieve this code and execute the resulting code containing instructions, setting the required values ​​for the variables in the code.

[0066] If the relevant background information included in the request for a generative AI model includes metadata of resources related to the input to the application, the request may, on the device on which the application is executed, retrieve the results of accessing those resources and use those results in addition to the input to invoke the generative AI model. Alternatively, at least one of the retrieval of those results and the invocation may occur on the device on which the generative AI model is provided.

[0067] For example, if the resource is a document and the metadata is the location where the document is stored, the document can be downloaded from that location and at least a portion of the document can be included in the instructions included in the call to the generating AI model. Alternatively, if the resource is a person and the metadata is the person's contact information, a question related to user input can be generated, the question can be sent to the contact, the answer to the question can be obtained, and at least a portion of the answer can be included in the instructions included in the call to the generating AI model. The resource may also be an AI assistant other than the application used by the user.

[0068] Finally, the application provides the user with output using the response received from platform 130 (S208).

[0069] The above explanation was mainly from the perspective of the application server 120, but the explanation from the perspective of the context server 100 is as follows. First, the context server 100 is a device that provides background information and receives a request from the user terminal 110 or an application running on the application server 120 to acquire relevant background information from the background information pre-stored in the storage device 104, in accordance with the user's input (S401). The acquisition request includes an identifier for the user or the group to which the user belongs, and the context server 100 creates a portion of the elements of the background information associated with the identifier as relevant background information (S402). The manner in which the relevant background information is created is as described above. Then, the context server 100 transmits the created relevant background information to the application (S403).

[0070] Furthermore, although the above description assumes that the context server 100 receives a request to acquire relevant background information from an application running on the user terminal 110 or the application server 120, the application may also run on the context server 100. In this case, the context server 100 should acquire the request rather than receive it from the application running on it.

[0071] Furthermore, in the embodiments described above, unless the word "only" is used, such as "based only," "depending only," "in the case of only," or "referencing only," it is assumed in this specification that additional information may also be considered. Also, as an example, the statement "if a, then b" does not necessarily mean "always b in the case of a" or "b immediately after a," unless explicitly stated otherwise. In addition, the statement "each a constituting A" does not necessarily mean that A is composed of multiple components, but includes the possibility that the component is singular.

[0072] Furthermore, it should be noted that the embodiments of the present invention described above are included in the disclosure herein, in any way that they are not inconsistent with each other.

[0073] Furthermore, for the sake of clarity, even if there are aspects of operation in some method, program, terminal, device, server, or system (hereinafter referred to as "method, etc.") that differ from the operation described herein, each aspect of the present invention is an invention that targets one of the operations described herein, and the existence of operation different from the operation described herein does not mean that the method, etc. is outside the scope of each aspect of the present invention.

[0074] Furthermore, the "start" and "end" shown in Figure 4 are merely examples and do not necessarily mean that the method according to this embodiment will always start or end in the illustrated procedure. [Explanation of symbols]

[0075] 100 context servers 101 Communications Department 102 Processing Unit 103 Storage section 104 Storage device 104-1 The first database 104-2 Second Database 110 user terminals 120 Application Servers 130 platforms

Claims

1. A method for causing a generative AI model to reflect unlearned information in its response to user input, The application obtains the input from the user, The application makes a request to acquire relevant background information corresponding to the input, The steps include: the application obtaining some or all of the elements of a predetermined format of background information associated with the application as the associated background information; The application makes a request to the generative AI model that includes the input and the related background information, The application provides the user with an output using the response to the request. Methods that include...

2. The method according to claim 1, A method wherein each of the aforementioned elements has one or more attributes.

3. A method according to claim 1 or 2, A method wherein a first element included in the plurality of elements is linked to a second element included in the plurality of elements.

4. A method according to claim 1 or 2, A method wherein the aforementioned related background information includes a portion of the plurality of elements that have a relatively high relevance to the input.

5. The method according to claim 4, The method further comprises one or more elements having a link with at least one of the aforementioned related background information.

6. A method for causing a generative AI model to reflect unlearned information in its response to user input, The application obtains the input from the user, The application makes a request to acquire relevant background information corresponding to the input, The steps include: the application obtaining some or all of a plurality of elements having background information of a predetermined format associated with the user or the application as associated background information, wherein the associated background information includes metadata of a resource related to the input; The steps include: the application accessing the resource using the metadata; The application makes a request to the generative AI model that includes the input and the results of the access, The application provides the user with an output using the response to the request. Methods that include...

7. The method according to claim 6, A method wherein the metadata includes metadata of an AI assistant related to the input.

8. A program for causing a computer to execute a method for reflecting unlearned information in the response of a generative AI model in response to user input, wherein the method is: The application obtains the input from the user, The application makes a request to acquire relevant background information corresponding to the input, The steps include: the application obtaining some or all of the elements of a predetermined format of background information associated with the application as the associated background information; The application makes a request to the generative AI model that includes the input and the related background information, The application provides the user with an output using the response to the request. A program that includes this.

9. A program for causing a computer to execute a method for reflecting unlearned information in the response of a generative AI model in response to user input, wherein the method is: The application obtains the input from the user, The application makes a request to acquire relevant background information corresponding to the input, The application obtains some or all of a plurality of elements having background information of a predetermined format associated with the user or the application as associated background information, wherein the associated background information includes metadata of a resource related to the input. The steps include: the application accessing the resource using the metadata; The application makes a request to the generative AI model that includes the input and the results of the access, The application provides the user with an output using the response to the request. A program that includes this.

10. A device for allowing a generative AI model to reflect unlearned information in its response to user input, wherein an application executed on the device is The system obtains the input from the user and requests the acquisition of relevant background information corresponding to the input or the user making the input. The system acquires some or all of the elements of background information in a predetermined format associated with the aforementioned application as the associated background information. A device configured to make a request to the generating AI model, including the aforementioned input and the aforementioned related background information, and to provide the user with an output using the aforementioned response to the request.

11. A device for a generating AI model to reflect unlearned information in its response to user input, wherein the application executed on the device is The system obtains the input from the user and requests the acquisition of relevant background information corresponding to the input or the user making the input. Some or all of the elements of a predetermined format of background information associated with the user or the application are acquired as related background information, which includes metadata of the resource related to the input, and the resource is accessed using the metadata. A device configured to make a request to the generating AI model including the aforementioned input and the results of the access, and to provide the user with an output using the response to the request.

12. The apparatus according to claim 10 or 11, A user terminal capable of communicating with the aforementioned device, wherein the user transmits the input and receives the output. A system equipped with these features.

Images