Generating large language model prompts based on textual user input for controlling lighting system

Abstract

A method of generating a hint of a large language model includes receiving (101) a textual user input for controlling a lighting system, obtaining (103) contextual information indicative of one or more contextual characteristics of a context of the user, a subset of configuration data related to the context of the user and / or an example subset of lighting control behaviors related to the context of the user is selected (105) based on the contextual characteristic. The method further includes generating (107) a hint of the large language model and communicating (109) the hint to the large language model. The prompt includes the textual user input and the selected one or more subsets. The prompt excludes configuration data that is not selected a portion of the configuration data subset, and excludes examples that are not selected a portion of the example subset.

Description

Technical Field

[0002] This invention relates to a system for generating prompts for large language models.

[0003] The present invention further relates to a method for generating prompts for a large language model.

[0004] The present invention also relates to a computer program product that enables a computer system to execute this method. Background Technology

[0006] Large language models such as GPT-4 can not only be used to provide chatbots like ChatGPT, but also allow users to control devices using natural language. Systems such as Google Assistant and Amazon Alexa already allow users to control devices, such as lighting equipment from the Hue lighting system, but only for specific commands.

[0007] Large language models can be used to control devices without requiring specific user commands (e.g., "I feel happy and like I'm going to fly."). However, this requires appropriate prompts for the large language model. Prompt engineering is the process of structuring text that a generative AI model can interpret and understand. Prompt engineering can consist of a single prompt that includes several examples from which the model will learn, a method known as few-shot learning. For large language models to be used economically, it is important that the generated prompts have an acceptable length.

[0008] The paper "Supporting Vision-Language Model Inference with Causality-pruning Knowledge Prompt" (arXiv:2205.11100) by Jiangmeng Li et al. describes a method for automatically generating prompts. This paper recognizes the importance of including semantic information in the prompts and proposes Causality-pruned Knowledge Prompt (CapKP) to adapt pre-trained vision-language models for downstream image recognition. CapKP retrieves ontology knowledge graphs by treating text labels as queries for exploring task-relevant semantic information. To further refine the derived semantic information, CapKP introduces causal pruning by following the first principles of Granger causality. Summary of the Invention

[0010] The first objective of this invention is to provide a system that can automatically generate appropriate prompts of acceptable length for a large language model based on textual user input for controlling a lighting system.

[0011] A second objective of the present invention is to provide a method that can be used to automatically generate appropriate prompts of acceptable length for a large language model based on textual user input for controlling a lighting system.

[0012] In a first aspect of the invention, a system for generating a prompt for a large language model includes: at least one input interface, at least one transmitter, and at least one processor configured to: receive text user input for controlling a lighting system via the at least one input interface, the text user input being provided by a user; obtain context information indicating one or more contextual characteristics of the user's context; select, based on the one or more contextual characteristics, at least one of a subset of configuration data of the lighting system and a subset of example lighting control behaviors of the lighting system, the subset of configuration data being configuration data related to the user's context, and the subset of example behaviors being examples related to the user's context; generate the prompt for the large language model, the prompt including at least one of the subset of configuration data and the subset of example behaviors, and the text user input; if the subset of configuration data has been selected, exclude configuration data that is not part of the subset of configuration data, and if the subset of example behaviors has been selected, exclude examples that are not part of the subset of example behaviors; and transmit the prompt to the large language model via the at least one transmitter.

[0013] Enhancing textual user input by incorporating configuration data and / or examples of lighting control behavior of the lighting system in the prompt can make the response from a large language model more useful. Prompts of acceptable length can be generated by including only a subset of configuration data and / or a subset of examples in the prompt. The usefulness of the response from a large language model is not significantly affected, even with shorter prompts, by selecting a subset based on one or more contextual features of the user's context (e.g., the user's current activity).

[0014] For example, the text user input can be obtained by acquiring text, images, and / or audio. For instance, the text user input can be obtained in the form of a verbal prompt (speech). The text user input can be included in the prompt in the same or different form as it was acquired. As an example of the latter, the text user input can be obtained in the form of audio and then included in the prompt in the form of text or an image.

[0015] The at least one processor can be configured to receive a response to the prompt from the large language model via the at least one input interface, and, if the response indicates a lighting control command for the lighting equipment of the lighting system, control the lighting equipment according to the lighting control command. The text user input can indicate a desired light setting for the lighting system. Examples of lighting control behavior may include examples of text user input and corresponding desired lighting control commands.

[0016] For example, if the response includes a lighting control command or contains wording indirectly linked to a lighting control command, the response indicates the lighting control command. As an example of the latter, the response could indicate one or more colors and / or indicate a desired change in lighting. In this case, additional steps may be required to translate the response into an actual lighting control command.

[0017] The contextual information may include environmental information indicating the characteristics of the user's environment, and / or environmental information indicating the user's activities, as well as user preferences. For example, when a user is in the living room, the prompts submitted to the large language model (the API) may primarily or solely include living room-related configuration data. Living room-related examples may still be included, as these examples can also be applied to other rooms.

[0018] Characteristics of the user's environment may include one or more of the following: the user's current location (e.g., a room), the number of users in the same location, the temperature at the user's current location, and the light level at the user's current location. For example, if it is hot at the user's current location, examples with dynamic fireplace lighting effects can be excluded from the prompt.

[0019] For example, the at least one processor may be configured to select a subset of configuration data based on the one or more context characteristics by selecting configuration data relating to one or more lighting devices corresponding to the context characteristics, one or more rooms corresponding to the context characteristics, and / or one or more light scenes corresponding to the one or more context characteristics. For example, if the user is in a living room, configuration data relating to the group "living room" or to the lighting devices and light scenes in that group may be included in the prompt.

[0020] The at least one processor may be configured to select a subset of examples based on the one or more contextual characteristics by selecting examples of the user's activity corresponding to the one or more contextual characteristics. For example, the user's activity could be working or lounging. For example, if the contextual characteristic indicates that the user is working, examples that apply when the user is working can be selected (e.g., including "I can't read my document" as an example of text user input).

[0021] The at least one processor may be configured to: generate a new suggestion for the large language model if the response does not indicate a lighting control command, wherein the new suggestion includes at least a portion of the configuration data excluded from the suggestion if a first subset is selected, and the new suggestion includes at least a portion of the examples excluded from the suggestion if a second subset is selected, and transmit the new suggestion to the large language model via the at least one transmitter. The goal of this system is typically to obtain a lighting control command. Expanding the suggestion may be helpful if the large language model does not respond with a lighting control command. For example, if the response does not include a lighting control command and does not include keywords that can be (locally) converted into a lighting control command, the response does not indicate a lighting control command.

[0022] The at least one processor may be configured to: if the response indicates a lighting control command for a lighting device of the lighting system, evaluate the applicability of the lighting control command; if the applicability of the lighting control command exceeds a threshold, control the lighting device according to the lighting control command; and if the applicability of the lighting control command does not exceed the threshold, generate additional prompts for the large language model, the additional prompts including supplementary information, and transmit the additional prompts to the large language model via the at least one transmitter. For example, if the light settings specified in the lighting control command cannot be (fully) rendered or are not sustainable, the lighting control command may be considered inappropriate. In this case, extended prompts may be helpful.

[0023] The potential spatial distance between the light settings specified in the lighting control command and the light settings that have proven useful can be used to indicate the applicability of the lighting control command. If the subset of configuration data is selected (and therefore not all configuration data is included in the prompt), the additional information may include a portion of the configuration data excluded from the prompt, and if the sample subset is selected (and therefore not all examples are included in the prompt), the additional information may include a portion of the examples excluded from the prompt.

[0024] The at least one processor can be configured to specify in the prompt a maximum permissible variation between standard lighting settings and one or more lighting settings to be generated by the large language model and included in the lighting control command in the response to the prompt. For example, the prompt may specify that the output of the decorative part of the lighting scene can have a higher degree of variation compared to the more functional lighting part of the lighting scene (e.g., the latter may have to adhere to predefined focus / reading / relaxation lighting settings).

[0025] The text user input may include at least a portion of the context information, and the at least one processor may be configured to obtain the at least a portion of the context information from the text user input. For example, if the text user input includes "I am hot," the fact that the user is hot can be used as context information. In this case, examples with fireplace dynamic lighting effects can be excluded from the prompt.

[0026] The at least one processor can be configured to confirm a target lighting control function and / or a target location based on the text user input, further select the sample subset based on the target lighting control function and / or the target location, or select the sample subset of lighting control behaviors based on the target lighting control function and / or the target location, and generate the prompt of the large language model, such that the prompt includes the text user input and the sample subset. As a first example, if the text user input includes "Is anyone in the attic?", examples that are not useful in the attic can be excluded. As a second example, if the text user input includes "I don't like the lights in this room", examples related to sensors can be excluded from the prompt.

[0027] The at least one processor can be configured to select the subset of configuration data based on the one or more contextual features, select the subset of examples based on the target lighting control function and / or the target location, and generate the prompt of the large language model, such that the prompt includes the text user input, the subset of configuration data, and the subset of examples. For example, if the text user input includes "I don't like the lights in this room" and the user is in the living room, examples related to the sensor can be excluded from the prompt, and configuration data unrelated to the living room can also be excluded from the prompt.

[0028] At least one processor can be configured to select a subset of examples of lighting control behavior of a lighting system by generating examples based on one or more contextual characteristics, such as by using a local LLaMA model.

[0029] At least one processor can be configured to, after obtaining examples of the lighting control behavior of the lighting system, rewrite one or more of the examples and include the rewritten examples in the prompt to be passed to the large language model. For example, at least one processor can be configured to select a subset of examples of the lighting control behavior of the lighting system by obtaining examples of the lighting control behavior of the lighting system from memory based on one or more contextual features, and rewrite one or more of the obtained examples, for example, by using a local LLaMA model.

[0030] In a second aspect of the invention, a method for generating a prompt for a large language model includes: receiving text user input for controlling a lighting system, the text user input being provided by a user; obtaining context information indicating one or more contextual characteristics of the user's context; selecting, based on the one or more contextual characteristics, at least one of a subset of configuration data of the lighting system and a subset of example lighting control behaviors of the lighting system, the subset of configuration data being configuration data related to the user's context, and the subset of example behaviors being examples related to the user's context; generating the prompt for the large language model, the prompt including at least one of the subset of configuration data and the subset of example behaviors, and the text user input; if the subset of configuration data has been selected, excluding configuration data that is not part of the subset of configuration data; and if the subset of example behaviors has been selected, excluding examples that are not part of the subset of example behaviors; and transmitting the prompt to the large language model. The method can be performed by software running on a programmable device. The software can be provided as a computer program product.

[0031] In addition, a computer program for performing the methods described herein is provided, as well as a non-transitory computer-readable storage medium for storing the computer program. The computer program may be downloaded or uploaded to an existing device, for example, or stored during the manufacture of these systems.

[0032] A non-transitory computer-readable storage medium stores at least one portion of software code that, when executed or processed by a computer, is configured to perform executable operations for generating prompts for a large language model.

[0033] The executable operation includes: receiving text user input for controlling a lighting system, the text user input being provided by a user; obtaining context information indicating one or more contextual characteristics of the user's context; selecting at least one of a configuration data subset of the lighting system and an example subset of lighting control behaviors of the lighting system based on the one or more contextual characteristics, the configuration data subset being configuration data related to the user's context, and the example subset being examples related to the user's context; generating the prompt for the large language model, the prompt including at least one of the configuration data subset and the example subset, as well as the text user input; if the configuration data subset has been selected, excluding configuration data that is not part of the configuration data subset, and if the example subset has been selected, excluding examples that are not part of the example subset; and transmitting the prompt to the large language model.

[0034] As those skilled in the art will understand, aspects of the present invention can be embodied as an apparatus, method, or computer program product. Therefore, aspects of the present invention can take the form of a completely hardware embodiment, a completely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects, all of which are generally referred to herein as a “circuit,” “module,” or “system.” The functionality described in this disclosure can be implemented as an algorithm executed by a computer’s processor / microprocessor. Furthermore, aspects of the present invention can take the form of a computer program product embodied in one or more computer-readable media having embodied (e.g., stored thereon) computer-readable program code.

[0035] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing, but is not limited thereto. More specific examples of computer-readable storage media may include, but are not limited to, the following: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing. In the context of this invention, a computer-readable storage medium can be any tangible medium that can contain or store a program used by or in connection with an instruction execution system, apparatus, or device.

[0036] Computer-readable signal media may include propagated data signals embodying computer-readable program code therein (e.g., in baseband or as part of a carrier wave). Such propagated signals may take any of a variety of forms, including but not limited to electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that is not a computer-readable storage medium and may communicate, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device.

[0037] Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, fiber optic, cable, RF, etc., or any suitable combination thereof. Computer program code used to perform operations of various aspects of this invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java™, Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0038] The following description refers to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, particularly a microprocessor or central processing unit (CPU), to produce a machine such that the instructions, executable via the processor of the computer, other programmable data processing apparatus, or other device, create means for implementing the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams.

[0039] These computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing apparatus or other device to operate in a particular manner, such that the instructions stored in the computer-readable medium produce an article of writing which includes instructions that implement the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0040] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide for implementing the functions / actions specified in one or more boxes of the flowchart and / or block diagram.

[0041] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, including one or more executable instructions for implementing a specified logical function(s). It should also be noted that in some alternative implementations, the functions indicated in the blocks may not occur in the order shown in the figures. For example, depending on the functions involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system performing the specified function or action, or by a combination of dedicated hardware and computer instructions. Attached Figure Description

[0043] Referring to the accompanying drawings, these and other aspects of the invention will be clear and further illustrated by way of example, in which: Figure 1 This is a block diagram of an embodiment of the system; Figure 2 This is a flowchart of the first embodiment of the method; Figure 3 This is a flowchart of the second embodiment of the method; Figure 4 This is a flowchart of the third embodiment of the method; Figure 5 This is a flowchart of the fourth embodiment of the method; Figure 6 This is a flowchart of the fifth embodiment of the method; Figure 7 This is a flowchart of the sixth embodiment of the method; and Figure 8 This is a block diagram of an exemplary data processing system for performing the methods of the present invention.

[0044] Corresponding elements in the accompanying drawings are indicated by the same reference numerals. Detailed Implementation

[0046] Figure 1 An embodiment of a system for generating prompts for a large language model is shown. In this embodiment, the system is a computer 21. Computer 21 is connected to the Internet 11 and acts as a server. For example, computer 21 could be operated by a lighting company. Figure 1In one example, the user can interact with mobile device 41 via voice and / or via a (virtual) keyboard. For example, mobile device 41 could be a mobile phone or a tablet. In another example, the user can interact with a smart speaker system via voice.

[0047] Voice input received by mobile device 41 is converted into text. This text, or text typed by the user, is transmitted to computer 21 as text user input. Computer 21 communicates with another internet server (or cluster of internet servers) hosting a large language model 13 (e.g., GPT-4). Computer 21 generates prompts for the large language model 13 based on the text user input and then receives a response from the large language model 13. This response is parsed, and (part of) the response can be forwarded to the user via mobile device 41.

[0048] exist Figure 1 In the example, the user's lighting system 21 includes a bridge 16 and lighting fixtures 31-34. Optionally, computer 1 (or a portion thereof) is also considered part of the lighting system 21. Computer 1 is configured to control the illumination of lighting fixtures 31-34 via bridge 19. Bridge 19 communicates with lighting fixtures 31-34, for example, using Zigbee technology. Alternatively or additionally, computer 1 may be able to control one or more of lighting fixtures 31-34 without a bridge.

[0049] Bridge 19 is connected to wireless LAN access point 17, for example, via WiFi or Ethernet. Wireless LAN access point 17 is connected to the Internet 11. Figure 1 In this example, mobile device 41 connects to the Internet via wireless LAN access point 17. Alternatively, mobile device 41 may connect to the Internet 11 via, for example, a 5G or 6G cellular network.

[0050] In an alternative embodiment, the mobile device 41 generates a prompt, transmits the prompt to the large language model 13, receives a response from the large language model 13, parses the response, and controls the lighting devices 31-34, for example, via the bridge 16.

[0051] Computer 1 includes receiver 3, transmitter 4, processor 5, and storage device 7. Processor 5 is configured to receive text user input for controlling lighting system 21 via receiver 3. This text user input is provided by a user, for example, via mobile device 41. Processor 5 is also configured to obtain context information indicating one or more context characteristics of the user's context, and to select at least one subset of configuration data for lighting system 21 and an example subset of lighting control behaviors of lighting system 21 from storage device 7 based on these one or more context characteristics. For example, the examples may be stored in storage device 7. For example, configuration data may be stored in storage device 7 or the memory of bridge 16.

[0052] For example, some or all of the contextual information can be obtained from text user input. Some or all of the contextual user input can be obtained from mobile device 41, bridge 16, and / or other devices (e.g., cameras or other sensors) in the same building. As a first example, it can be obtained from a camera ( Figure 1 (Not shown) Information about the user's activities, the user's location, and / or the number of people present at the same location as the user can be received from a temperature sensor ( Figure 1 (Not shown) Receives information about the temperature at the user's location. The selected subset of configuration data is context-dependent configuration data for the user. The selected subset of examples is context-dependent examples for the user.

[0053] Processor 5 is further configured to generate a prompt for the large language model 13. The prompt includes textual user input and at least one subset of configuration data and a subset of examples. If a subset of configuration data has been selected, the prompt excludes configuration data that is not part of the configuration data subset, and if a subset of examples has been selected, the prompt excludes examples that are not part of the example subset. The textual user input can be included in the prompt in the same or different form as it was obtained. For clarity, the textual user input can be preprocessed before being included in the prompt.

[0054] Processor 5 is further configured to transmit the prompt to the large language model 13 via transmitter 4. The prompt is transmitted to the large language model 13 by transmitting the prompt to an Internet server (or cluster of Internet servers) hosting the large language model.

[0055] exist Figure 1In one embodiment, the processor is configured to receive a response to the prompt from the large language model 13 via receiver 3, and if the response indicates (e.g., includes) a lighting control command for the lighting devices of the lighting system, control one or more of the lighting devices 31-34 according to the lighting control command. The processor 5 may be configured to generate a new prompt if the response does not indicate (e.g., does not include) a lighting control command, and transmit the new prompt to the large language model 13.

[0056] Processor 5 can be configured to evaluate the applicability of the lighting control command included in the response or the applicability of the entire response, and if the applicability does not exceed a threshold, generate a new prompt and pass the new prompt to the large language model 13.

[0057] Processor 5 can be configured to generate examples of lighting control behaviors of the lighting system based on the user's context. Additionally or alternatively, processor 5 can be configured to rewrite one or more of these examples after obtaining them from storage device 7, and include the rewritten examples in the prompt to be sent to the large language model 13. For example, they can be rewritten by combining two examples obtained from storage device 7 into a single example, thereby creating a new example that includes a first part of the first example and a second part of the second example. This can be done to compress elaborate examples into more concise and clear examples. While this may result in the loss of some information from the stored examples, it will reduce the cost of cloud API calls to the large language model 13. Examples can also be generated or rewritten to give them different tones, such as being more empathetic, stricter, or offering more leeway in the wording. For example, examples can be generated or rewritten by a local LLaMA model running on processor 5.

[0058] exist Figure 1 In the embodiment of computer 1 shown, computer 1 includes a processor 5. In alternative embodiments, computer 1 includes multiple processors. The processor 5 of computer 1 may be a general-purpose processor (e.g., from Intel or AMD) or a dedicated processor. For example, the processor 5 of computer 1 may run a Windows- or Unix-based operating system. Storage device 7 may include one or more memory units. For example, storage device 7 may include one or more hard disks and / or solid-state drives. For example, storage device 7 may be used to store operating systems, applications, and application data.

[0059] For example, receiver 3 and transmitter 4 can communicate with devices on the Internet 11 using one or more wired and / or wireless communication technologies such as Ethernet and / or Wi-Fi (IEEE 802.11). In alternative embodiments, multiple receivers and / or multiple transmitters are used instead of a single receiver and a single transmitter. Figure 1 In the illustrated embodiment, a separate receiver and a separate transmitter are used. In an alternative embodiment, receiver 3 and transmitter 4 are combined into a transceiver. Computer 1 may include other components typical of a computer, such as a power connector. The invention can be implemented using a computer program running on one or more processors.

[0060] exist Figure 1 In one embodiment, the system of the present invention is an internet server. In an alternative embodiment, the system can be another device, such as a mobile device. Figure 1 In one embodiment, the system of the present invention includes a single device. In an alternative embodiment, the system of the present invention includes multiple devices, such as a cloud computing cluster or computer 1 and mobile device 41.

[0061] Figure 2 The first embodiment of a method for generating prompts for a Large Language Model (LLM) is shown. For example, the method can be derived from... Figure 2 The computer 1 executes the procedure. Step 101 includes receiving text user input for controlling the lighting system. The text user input is provided by the user. For example, the text user input may indicate the desired light setting for the lighting system. Examples of text user input are “Make the office forest green,” “I’m leaving the house in ten seconds,” “I’m back,” and “I’m going to the kitchen.”

[0062] Step 103 includes obtaining contextual information that indicates one or more contextual characteristics of the user's context. Contextual information may include, for example, one or more of the following: environmental information indicating characteristics of the user's environment, environmental information indicating the user's activities, and user preferences. Characteristics of the user's environment may include one or more of the following: the user's current location (e.g., a room), the number of users present at the same location, the temperature at the user's current location, and the light level at the user's current location.

[0063] Step 105 includes selecting a subset of configuration data for the lighting system and / or a subset of examples of lighting control behaviors based on one or more contextual characteristics obtained in step 103. The subset of configuration data (if selected) is context-dependent configuration data related to the user. For example, if historical data from the lighting system indicates which lighting scenarios have been (frequently) selected by the user previously, these scenarios can be included in the prompt. Therefore, configuration data is selected based on user preferences. The subset of examples (if selected) is context-dependent examples related to the user. Multiple examples of lighting control behaviors may each include examples of textual user input and corresponding desired lighting control commands.

[0064] Step 107 includes a prompt to generate a large language model. This prompt includes the text user input received in step 101 plus the subset of configuration data selected in step 105 and / or the subset of examples selected in step 105. If a subset of configuration data has been selected in step 105, the prompt excludes configuration data that is not part of the subset, and if a subset of examples has been selected in step 105, the prompt excludes examples that are not part of the subset. If no subset of examples has been selected in step 105, all examples can be included in the prompt. Similarly, if no subset of configuration data has been selected in step 105, all configuration data can be included in the prompt.

[0065] The prompt may also include one or more of the following: • The internal knowledge and information of a large language model may only be up-to-date at a certain point in time. For example, GPT-4's internal knowledge and information are only up-to-date at a certain point in 2021. If the text user input involves something that happened recently (e.g., the user mentions that they liked the stage lighting during the recent Super Bowl halftime show), a prompt can specify that the large language model should perform a web search to help update the large language model's knowledge to the Super Bowl 2023 halftime show.

[0066] • When generating dynamic lighting content such as entertainment lighting effects, hints can instruct the large language model to rely on its own knowledge to generate lighting settings without resorting to (potentially copyrighted) content from online sources.

[0067] • If a user requests a light setting that is harmful to someone's health (e.g., a user requests a flashlight formula that could harm someone with epilepsy or says "my mom had a terrible night's sleep tonight"), the prompt can ask the large language model to create a disclaimer indicating that the ultimate responsibility for control over the lighting equipment lies with the user (alternatively, the user's request can be denied).

[0068] The prompt can specify one or more parameter values ​​for the large language model. One or more of these parameter values, such as the value of the "temperature" parameter, can depend on contextual information, such as the user's current location. The value of the parameter called "temperature" defines the degree of determinism in the large language model's response. For example, the "temperature" value can be increased for rooms where people might prefer to experience more variable lighting settings (e.g., living rooms, man caves), and decreased for other rooms (e.g., home offices) to make the response more deterministic. Step 109 includes passing the prompt generated in step 107 to the large language model. Additionally, the prompt can be... Figure 3-7 One or more steps of one or more embodiments are added to Figure 2 Examples of implementations.

[0069] Figure 3 The second embodiment of a method for generating prompts for a large language model is shown in the figure. Figure 3 The embodiment is Figure 2 An extension of the embodiments.

[0070] exist Figure 3 In the embodiments, in Figure 2 Steps 121 and 123 are executed after step 109. Step 121 includes receiving a response from the large language model to the prompt transmitted in step 109. Step 123 includes checking whether the response received in step 121 indicates (e.g., includes) a lighting control command for the lighting equipment of the lighting system. If so, step 125 can be executed. Step 125 includes controlling the lighting equipment according to the lighting control command indicated (e.g., included) in the response received in step 121. Approximately while controlling the lighting equipment, a portion of the response from the large language model can be forwarded to the user. This can be done, for example, in step 125 or a different step.

[0071] Optionally, step 125 is performed only if certain conditions are met. For example, if step 123 determines that the response received in step 121 indicates (e.g., includes) a lighting control command for a lighting device of the lighting system, step 137 can be performed first. Step 137 includes evaluating the applicability of the lighting control command indicated (e.g., included) in the response received in step 121. Next, step 139 includes checking whether the applicability of the lighting control command exceeds a threshold. If so, step 125 is performed, and a portion of the response received in step 121 can be forwarded to the user.

[0072] For example, the suitability of lighting control commands can be evaluated using one or more of the following methods: • Part of the applicability assessment can be derived from the large language model itself. If the response from the large language model indicates that the large language model has low confidence in its response (e.g., by returning multiple drafts of the answer for the user to choose from instead of a single answer, or by directly activating the light without any intervention), the lighting control command can be considered inapplicable.

[0073] • Is the response received from the large language model operable (e.g., does it change any meaningful lighting settings; is the palette proposed by the large language model within the color gamut of the lighting system; can the proposed colors be rendered, e.g., not brown or gray light; can dynamic effects be rendered given system constraints)? For example, the potential spatial distance between the lighting settings specified in a lighting control command and the lighting settings that have proven useful can be used to indicate the applicability of the lighting control command.

[0074] • Is the lighting setup specified in the response from the large language model sufficiently sustainable (e.g., predefined sustainability targets for the lighting system) (e.g., overall energy consumption, carbon intensity, and lifespan degradation of the luminaires)? • How visible will the proposed changes to the lighting settings be to the user? For example, are the lighting settings specified in the response from the large language model so subtle that they are almost invisible, or are they too intrusive (e.g., using a flash during the user's weaving process)? How appealing / pleasant / logical / surprising will the lighting settings specified in the response from the large language model be to the user? • Is the lighting setting specified in the response from the large language model logical for the user? For example, is it logical to present lighting effects in the basement when the user is having a party in the outdoor garden? • Is the logic and reasoning of the large language model for the proposed lighting control commands rigorous, intelligent, and defensible (e.g., can the proposed light settings be interpreted if the user requests an explanation)? • Does the response from the large language model include lighting control commands that the user or lighting system cannot or is not authorized to execute? For example, does the large language model ask the user to turn off a street light or turn on a neighbor's light? • Do responses from the large language model include offensive lighting control commands? For example, do lighting control commands correspond to lighting effects specific to a 50-year-old, such as dementia treatment lighting?

[0075] • What impact do lighting settings have on user well-being? • Regarding light settings that affect user well-being, has the large language model performed any web searches and do the results from those searches support the reasoning behind the proposed light settings? The large language model is prone to illusions and may construct seemingly highly reasonable but scientifically flawed light settings, for example, for improving user sleep.

[0076] • For light settings that affect human health (e.g., UV lighting, circadian rhythm lighting), does the large language model only use facts from search results or also add AI-generated information itself?

[0077] The applicability assessment can be performed by another AI model that evaluates the quality of the logical reasoning steps (e.g., thought chain steps) used by the large language model to obtain its inferences.

[0078] If it is determined in step 139 that the applicability of the lighting control command does not exceed a threshold, steps 141 and 143 are performed. Step 141 includes generating additional prompts for the large language model. These additional prompts include supplementary information. If a subset of configuration data was selected in step 105, the supplementary information may include a portion of the configuration data excluded from the prompts. If a subset of examples was selected in step 105, the supplementary information may include a portion of the examples excluded from the prompts.

[0079] This additional information may include instructions for the large language model, such as how closely it needs to conform to the example or to adaptively guide the large language model to different predefined decision-making branches (e.g., predefined inference branches given in the example or predefined inference branches that the large language model sees in the training data during its training).

[0080] This additional information may include information from the user. For example, if the applicability of a lighting control command is assessed as low, the user may be asked to provide an image, such as an image of the user's bedroom, which may be included in additional prompts within the larger language model.

[0081] If the light setting specified in the response from the large language model affects the user's well-being, the additional hint can specify how many high-quality web search results are needed as evidence that the large language model's reasoning for that light setting is factual, depending on the lighting use case (e.g., more results can be requested for UV light settings than for ambient lighting effects).

[0082] If the light settings specified in the response from the large language model affect human health (e.g., UV lighting, circadian rhythm lighting), the additional prompt can require the large language model to use only facts from the search results and not to add any AI-generated information on its own. For example, when the additional response to the additional prompt is forwarded to the user, the user can be asked to make a final decision.

[0083] If the response from the large language model includes an offensive lighting control command, the additional cue can be expanded using additional descriptions from the user. For example, if the lighting control command corresponds to a lighting effect specific to a 50-year-old, such as dementia treatment lighting, the cue could be expanded to describe the user as a highly active 50-year-old. Step 143 involves passing the additional cue to the large language model. Step 121 is repeated after step 143, and the method then proceeds as follows: Figure 3 The process continues as shown.

[0084] Optionally, if it is determined in step 123 that the response received in step 121 does not indicate (e.g., does not include) a lighting control command for the lighting devices of the lighting system, steps 133 and 135 are performed. Step 133 includes generating a new prompt for the large language model. In step 105, if a first subset is selected, the new prompt includes at least a portion of the configuration data excluded from the prompt, and if a second subset is selected, the new prompt includes at least a portion of the examples excluded from the prompt. Step 135 includes transmitting the new prompt to the large language model. Step 121 is repeated after step 135, and the method then proceeds as follows. Figure 3 The process continues as shown.

[0085] Optionally, step 107 includes step 131. Step 131 includes specifying in the prompt the maximum permissible variation between a standard lighting setting and one or more lighting settings to be generated by the large language model and included in the lighting control command in response to the prompt. For example, a single decorative lighting scene may include functional and decorative portions. In this case, the prompt may request the large language model to adjust the lighting scene only with respect to the functional or decorative portion. For instance, the prompt may specify that the output of the decorative portion of the scene can have a higher degree of variation compared to the more functional lighting portion of the scene (e.g., the latter may have to adhere to predefined focus / reading / relaxation lighting settings). Figure 2 As described, this can also be achieved by changing the model parameters of a large language model (e.g., "temperature").

[0086] Additionally or alternatively, this change may depend on one or more contextual characteristics obtained in step 103. For example, if the user's activity requires more functional light (e.g., working), the cue may specify a lower maximum change, and if the user's activity requires less functional light (e.g., lounging), the cue may specify a higher maximum change, even if the user actually uses the same verbal cue text (e.g., "Let the light relax me") in both the lounging and working situations. Additionally, the cue may be... Figure 4-7 One or more steps in the embodiments are added to Figure 3 In the embodiments described above.

[0087] In an alternative embodiment, step 137 further includes checking the applicability of other portions of the response from the large language model, for example, in the following manner: • Does the response from the large language model include a brief suggestion based on additional text user input? If so, the lighting control command can be considered appropriate, the lighting equipment can be controlled according to the lighting control command, and the brief suggestion can be provided to the user. Therefore, because the final choice of lighting effect may not be made until after the next text user input, the user is exposed to an initial imperfect response from the large language model.

[0088] In this alternative embodiment, step 139 includes checking whether the applicability of the entire response exceeds a threshold. If so, step 125 is performed, and a portion of the response received in step 121 can be forwarded to the user.

[0089] Figure 4 The image shows a third embodiment of a method for generating prompts for a large language model. For example, this method can be derived from... Figure 2 The computer 1 executes the procedure. Step 101 includes receiving text user input for controlling the lighting system. This text user input is provided by the user. Step 103 includes obtaining context information indicating one or more contextual characteristics that suggest the user's context.

[0090] exist Figure 4 In this embodiment, the text user input received in step 101 includes at least a portion of contextual information, and step 103 includes sub-step 161. Step 161 includes obtaining at least a portion of the contextual information from the text user input received in step 101. For example, if the text user input includes "I am hot," the fact that the user is hot can be used as contextual information. In this case, examples with fireplace dynamic lighting effects can be excluded from the prompt.

[0091] Step 105 includes selecting a subset of the lighting system's configuration data and / or an example subset of the lighting system's lighting control behaviors based on one or more contextual characteristics obtained in step 103. The configuration data subset is context-dependent configuration data related to the user. The example subset is context-dependent examples related to the user.

[0092] Step 107 includes a prompt to generate a large language model. This prompt includes the textual user input received in step 101, plus the subset of configuration data selected in step 105 and / or the subset of examples selected in step 105. If a subset of configuration data has already been selected in step 105, the prompt excludes configuration data that is not part of that subset; and if a subset of examples has already been selected in step 105, the prompt excludes examples that are not part of that subset.

[0093] Step 109 includes transmitting the prompt generated in step 107 to the large language model. Additionally, it can be... Figure 3 and Figure 5-7 One or more steps in the embodiments are added to Figure 4 Examples of implementations.

[0094] Figure 5 The fourth embodiment of a method for generating prompts for a large language model is shown in the figure. Figure 5 The embodiment is Figure 2 An extension of the embodiments. In Figure 5 In the embodiments, Figure 2 Step 107 includes sub-step 171 and / or sub-step 173.

[0095] Step 171 includes selecting a subset of configuration data based on one or more contextual characteristics by selecting configuration data associated with at least one of the following: one or more lighting devices corresponding to the contextual characteristics obtained in step 103, one or more rooms corresponding to the contextual characteristics obtained in step 103, and one or more light scenes corresponding to the one or more contextual characteristics obtained in step 103.

[0096] Step 173 includes selecting a subset of examples based on one or more contextual characteristics by selecting examples of user activity corresponding to one or more contextual characteristics obtained in step 103. Additionally, it can be... Figure 3-4 and Figure 6-7 One or more steps in the embodiments are added to Figure 5 Examples of implementations.

[0097] Figure 6 The fifth embodiment of a method for generating prompts for a large language model is shown. For example, the method can be derived from... Figure 2The computer 1 executes.

[0098] Step 101 includes receiving text user input for controlling the lighting system. This text user input is provided by the user. After step 101 is completed, step 181 is executed. Step 181 includes confirming the target lighting control function and / or target location based on the text user input.

[0099] Step 103 includes obtaining context information that indicates one or more contextual characteristics of the user's context. For example, step 103 may be performed after step 101, such as if the context information is obtained from the text user input, or it may be performed in parallel with step 101. Step 183 is performed after steps 181 and 103 have been completed.

[0100] Step 183 includes selecting an example subset of lighting control behaviors of the lighting system based on the target lighting control function and / or target location identified in step 181. As a first example, if the text user input includes "Is anyone in the attic?", examples that are not useful in the attic can be excluded. As a second example, if the text user input includes "I don't like the lights in this room", examples related to sensors can be excluded from the prompt.

[0101] A subset of examples can be further selected based on one or more contextual characteristics obtained in step 103, such that the subset of examples is context-dependent on the user. Step 183 may further include selecting a subset of configuration data for the lighting system based on one or more contextual characteristics obtained in step 103, such that the subset of configuration data is context-dependent on the user. Therefore, any one or both of the subset of examples and the subset of configuration data are selected based on one or more contextual characteristics.

[0102] Step 107 includes generating a prompt for a large language model. This prompt includes at least the textual user input received in step 101 and the subset of examples selected in step 183. If a subset of configuration data for the lighting system was selected in step 183, that subset is also included in the prompt. Otherwise, all configuration data is included in the prompt. The prompt excludes examples that are not part of the subset of examples selected in step 181. If a subset of configuration data was selected in step 183, the prompt excludes configuration data that is not part of that subset.

[0103] Step 109 includes sending the prompt generated in step 107 to the large language model. Additionally, it can be... Figure 3-5 and Figure 7 One or more steps in the embodiments are added to Figure 6 Examples of implementations.

[0104] Figure 7 The sixth embodiment of a method for generating prompts for a large language model is shown in the figure. Figure 7 The embodiment is Figure 6 An extension of the embodiments. In Figure 7 In the embodiments, Figure 6 Step 183 includes steps 191 and 193.

[0105] Step 191 includes selecting a subset of configuration data based on one or more contextual characteristics obtained in step 103. Step 193 includes selecting a subset of examples based on the target lighting control function and / or target location confirmed in step 181. For example, if the text user input includes "I don't like the lights in this room" and the user is in the living room, examples related to the sensor can be excluded from the prompt, and configuration data not related to the living room can also be excluded from the prompt.

[0106] Step 107 includes generating a prompt for the large language model, such that the prompt includes the textual user input received in step 101, the subset of configuration data selected in step 191, and the subset of examples selected in step 193. Additionally, it can be... Figure 3-5 One or more steps in the embodiments are added to Figure 7 Examples of implementations.

[0107] Figure 8 The description shows that it can be performed as shown in the reference. Figure 2-7 A block diagram of an exemplary data processing system describing the method.

[0108] like Figure 8 As shown, the data processing system 300 may include at least one processor 302 coupled to a memory element 304 via a system bus 306. Thus, the data processing system can store program code within the memory element 304. Furthermore, the processor 302 can execute program code accessed from the memory element 304 via the system bus 306. In one aspect, the data processing system may be implemented as a computer suitable for storing and / or executing program code. However, it should be understood that the data processing system 300 may be implemented in the form of any system including a processor and memory capable of performing the functions described in this specification. For example, the data processing system may be an Internet / cloud server.

[0109] Memory element 304 may include one or more physical memory devices, such as, for example, local memory 308 and one or more mass storage devices 310. Local memory may refer to random access memory or (multiple) other non-persistent memory devices typically used during the actual execution of program code. Mass storage devices may be implemented as hard disk drives or other persistent data storage devices. Processing system 300 may also include one or more cache memories (not shown) that provide temporary storage for at least some program code to reduce the number of times program code must be retrieved from mass storage device 310 during execution. For example, if processing system 300 is part of a cloud computing platform, processing system 300 may also be able to use memory elements of another processing system.

[0110] Input / output (I / O) devices, depicted as input device 312 and output device 314, can be optionally coupled to the data processing system. Examples of input devices may include, but are not limited to, a keyboard, a pointing device such as a mouse, a microphone (e.g., for voice and / or speech recognition), etc. Examples of output devices may include, but are not limited to, a monitor or display, a speaker, etc. Input and / or output devices may be coupled to the data processing system directly or through an intermediate I / O controller.

[0111] In one embodiment, the input and output devices can be implemented as a combined input / output device (such as...). Figure 8 (Illustrated in the diagram using dashed lines surrounding input device 312 and output device 314). An example of such a combined device is a touch-sensitive display, sometimes also called a "touchscreen display" or simply a "touchscreen". In such embodiments, input to the device can be provided by the movement of a physical object, such as, for example, a stylus or a user's finger, above or near the touchscreen display.

[0112] Network adapter 316 can also be coupled to the data processing system to enable it to couple to other systems, computer systems, remote network devices, and / or remote storage devices via an intermediate private or public network. The network adapter may include a data receiver for receiving data transmitted to the data processing system 300 from the systems, devices, and / or networks, and a data transmitter for transmitting data from the data processing system 300 to the systems, devices, and / or networks. Modems, cable modems, and Ethernet cards are examples of different types of network adapters that can be used with the data processing system 300.

[0113] like Figure 8As depicted, memory element 304 can store application 318. In various embodiments, application 318 can be stored in local memory 308, one or more mass storage devices 310, or separately from local memory and mass storage devices. It should be understood that data processing system 300 can further execute an operating system (…). Figure 8 (Not shown in the image), the operating system can facilitate the execution of application 318. Application 318, implemented as executable program code, can be executed by data processing system 300, for example, by processor 302. In response to executing the application, data processing system 300 can be configured to perform one or more operational or method steps described herein.

[0114] Various embodiments of the present invention can be implemented as program products for use with a computer system, wherein the program(s) of the program product define the functionality of the embodiments (including the methods described herein). In one embodiment, the program(s) may be contained on a variety of non-transitory computer-readable storage media, wherein, as used herein, the expression “non-transitory computer-readable storage media” includes all computer-readable media, with the sole exception of transient propagation signals. In another embodiment, the program(s) may be contained on a variety of transient computer-readable storage media. Illustrative computer-readable storage media include, but are not limited to: (i) non-writable storage media (e.g., read-only memory devices within a computer, such as CD-ROM discs readable by a CD-ROM drive, ROM chips, and / or any type of solid-state non-volatile semiconductor memory) on which information is permanently stored; and (ii) writable storage media on which variable information is stored (e.g., flash memory, floppy disks or hard disk drives within a floppy disk drive, or any type of solid-state random access semiconductor memory). The computer program may run on the processor 302 described herein.

[0115] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well. It will be further understood that, when used in this specification, the terms “comprising” and / or “including” specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0116] All the means or steps plus functional elements in the following claims are intended to include any structure, material, action, and equivalent for performing a function in combination with other claimed elements specifically claimed. Descriptions of embodiments of the invention have been presented for illustrative purposes, but are not intended to be exhaustive or limited to the disclosed forms. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the invention. The embodiments have been chosen and described in order to best explain the principles of the invention and some practical applications, and to enable others skilled in the art to understand the invention with respect to various embodiments having various modifications suitable for the intended particular use.

Claims

1. A system (1) for generating prompts for a large language model (13), the system (1) comprising: At least one input interface (3). At least one transmitter (4); and At least one processor (5) is configured as follows: - Receive text user input for controlling the lighting system (21) via the at least one input interface (3), the text user input being provided by the user. - Obtain context information indicating one or more contextual characteristics of the user's context. - Based on the one or more contextual characteristics, select at least one of the following subsets: a subset of configuration data of the lighting system (21) and a subset of example lighting control behaviors of the lighting system (21), wherein the subset of configuration data is configuration data related to the user's context, and the subset of example behaviors is an example related to the user's context. - Generate the prompt for the large language model (13), the prompt including at least one of the configuration data subset and the example subset, and the text user input, wherein if the configuration data subset has been selected, the prompt excludes configuration data that is not part of the configuration data subset, and if the example subset has been selected, the prompt excludes examples that are not part of the example subset, and - The prompt is transmitted to the large language model (13) via the at least one transmitter (4).

2. The system (1) according to claim 1, wherein the at least one processor (5) is configured to: - Receive a response to the prompt from the large language model (13) via the at least one input interface (3), and - If the response indicates a lighting control command for the lighting equipment of the lighting system (21), control the lighting equipment according to the lighting control command.

3. The system (1) according to claim 2, wherein the at least one processor (5) is configured to: - If the response does not indicate a lighting control command, generate a new prompt for the large language model (13), which, if the first subset has been selected, includes at least a portion of the configuration data excluded from the prompt and / or, if the second subset has been selected, includes at least a portion of the example excluded from the prompt. - The new prompt is transmitted to the large language model (13) via the at least one transmitter (4).

4. The system (1) according to claim 2 or 3, wherein the at least one processor (5) is configured to: - If the response indicates a lighting control command for the lighting equipment of the lighting system (21), assess the applicability of the lighting control command. - If the applicability of the lighting control command exceeds a threshold, control the lighting device according to the lighting control command. - If the applicability of the lighting control command does not exceed the threshold, generate additional prompts for the large language model (13), the additional prompts including supplementary information, and - The additional prompt is transmitted to the large language model (13) via the at least one transmitter (4).

5. The system (1) of claim 4, wherein if the subset of configuration data is selected, the additional information includes at least a portion of the configuration data excluded from the prompt, and if the subset of examples is selected, the additional information includes at least a portion of the examples excluded from the prompt.

6. The system (1) according to any one of claims 2-5, wherein the at least one processor (5) is configured to specify in the prompt a maximum permissible variation between a standard light setting and one or more light settings to be generated by the large language model (13) to be included in the lighting control command in the response to the prompt.

7. The system (1) according to any one of the preceding claims, wherein the text user input includes at least a portion of the context information, and wherein the at least one processor (5) is configured to obtain the at least portion of the context information from the text user input.

8. The system (1) according to any one of the preceding claims, wherein the at least one processor (5) is configured to: - Confirm the target lighting control function and / or target location based on the user input of the text. - Further select the example subset based on the target lighting control function and / or the target location, or select the example subset of lighting control behaviors based on the target lighting control function and / or the target location, and - Generate the prompts for the large language model (13) such that the prompts include the text user input and the subset of examples.

9. The system (1) according to claim 8, wherein the at least one processor (5) is configured to: - Select the subset of configuration data based on one or more of the aforementioned contextual characteristics. - Select the sample subset based on the target lighting control function and / or the target location, and - Generate the prompts for the large language model (13) such that the prompts include the text user input, the subset of configuration data, and the subset of examples.

10. The system (1) according to any one of the preceding claims, wherein the context information includes at least one of the following: environmental information indicating the characteristics of the environment in which the user is located, environmental information indicating the activities of the user, and user preferences.

11. The system (1) according to any one of the preceding claims, wherein the at least one processor (5) is configured to select a subset of configuration data based on the one or more context features by selecting configuration data associated with at least one of the following: one or more lighting devices corresponding to the context feature, one or more rooms corresponding to the context feature, and one or more light scenes corresponding to the one or more context features.

12. The system (1) according to any one of the preceding claims, wherein the at least one processor (5) is configured to select the subset of examples based on the one or more context features by selecting examples of the user's activities corresponding to the one or more context features.

13. The system (1) according to any one of the preceding claims, wherein the plurality of examples of the text user input indicating the desired light setting and / or lighting control behavior for the lighting system (21) each includes an example of the text user input and a corresponding desired lighting control command.

14. A method for generating prompts for a large language model, the method comprising: - Receive (101) text user input for controlling the lighting system, the text user input being provided by the user; - Obtain (103) context information indicating one or more contextual characteristics of the user's context; - Based on the one or more context characteristics, select (105) at least one of the configuration data subset of the lighting system and the example subset of the lighting control behavior of the lighting system, wherein the configuration data subset is configuration data related to the user's context and the example subset is an example related to the user's context; - Generate (107) the prompt of the large language model, the prompt including at least one of the configuration data subset and the example subset and the text user input, the prompt excluding configuration data that is not part of the configuration data subset if the configuration data subset has been selected, and the prompt excluding examples that are not part of the example subset if the example subset has been selected; as well as - The prompt is transmitted (109) to the large language model.

15. A computer program product for a computing device, the computer program product comprising computer program code for executing the method of claim 14 when the computer program product is run on a processing unit of the computing device.

Images