A computer-implemented method for providing a response to a user query

Abstract

A computer-implemented method for providing a response consists of: receiving a user query from a user 10 by an integrated control system 20; generating an energy usage or pricing request based on the

Description

The present application relates to a computer-implemented method for providing a response to a user query. The growing usage of modern chatbot systems, particularly those based on generative artificial intelligence (AI) models, such as the Chat Generative Pre-trained Transformer (ChatGPT), has prompted the continuous refinement of these chatbots in order to enhance their capabilities and optimize the user experience. Generative AI models that employ AI techniques to comprehend and process human language text and generate human-like responses to user queries and inputs are generally known as Large Language Models (LLMs). With the ongoing progress of technology, the increased need for generative AI based on LLMs to extend beyond text-based interactions has led to the research, development and integration of multimodal features, allowing the users to interact with the generative AI also through an audio or video input. To achieve this, generative AI tools commonly utilize an LLM in combination with speech-to-text conversion models that are capable of converting the audio or the video's audio into text. To develop these functionalities, LLMs are trained on substantial amount of text, audio and video data, which requires considerable computational capacity and energy consumption that are expected to increase further over time. Alongside the environmental impact of LLMs, the increased preferences and expectations of users for improved quality, usefulness and personalization of the generated responses, have triggered the search for solutions to make the LLMs more economically viable, helpful and tailored to users. Typically, LLMs operate without access to personal user information and the responses they generate are not automatically customized to individual needs and preferences. Consequently, for the LLM to produce a more personalized response, the users are required to explicitly include all pertinent information in their queries which can be perceived as counterproductive by the users, lead to potential user frustration or have a limiting effect on the quality of the provided response. Specifically, attributes that can potentially improve the response quality may not be considered by the users. The quality of a generated LLM response may be influenced by various factors, including the quality of the query or input, the quantity and diversity of the data used for training the model, and how relevant that data is with respect to the query or input. Furthermore, training data, if not carefully preprocessed, may likely lead to the LLM learning incorrect information, showing biased behavior or experiencing what is known as AI hallucinations. Further, some LLMs operate with data collected from publicly available sources and do not use information from proprietary internet pages, real-time data or specific databases that often require payment or subscription for access. Typically, access to these LLMs is unrestricted. In general, an unrestricted utilization of an LLM may lead to excessive or irresponsible use of its computational resources, further contributing to the environmental footprint. LLMs can also be trained on data that includes copyrighted material and / or may require payment for generating responses to user queries. In general, a paid use of an LLM can offer various benefits, including improved service, advancements in functionality and continuous system development. Furthermore, a generated revenue may be invested, for example, in clean energy sources for powering the data centers that host the LLMs. Payment based LLMs, such as ChatGPT-4, primarily rely on subscription payments for using the LLMs' capabilities. However, subscription payments are prone to high churn rates or can discourage potential users, seeking more flexibility, from using the LLMs. LLMs use a significant amount of energy, and each query will use some amount of energy. It is often not known what the energy usage of a query will be before it the query is performed. Moreover, further training in the form of user feedback may in some instances reduce energy consumption in the future and so this may have an effect on the overall energy usage of the system over time. Moreover, as per above at some times of the day a system utilizing an LLM may be powered by renewable power (e.g. during daylight if powered by solar), whilst at other times there may be a high burden (e.g. at peak hours) which may cause system lag, or increase the heat produced and energy used by the system. These factors need to be taken into account when determining expected energy usage. In addition, subscription payments are fixed price payments and may be somewhat inequitable considering that different users have different needs and preferences or provide varying levels of detail and complexity to their queries or inputs. This way, a user seeking more straightforward information is charged the same as a person asking a more complex or personalized question. Further, according to reports from AI service providers, there is a huge demand for AI services which in many scenarios exceeds the available hardware resources. Thus, it is an objective of the present invention to provide an improved method for providing a response to a user query. In particular, there is a need to distribute available hardware resources appropriately to improve available service, e.g. to enhance responsiveness. In particular, it is an objective of the present application to provide an alternative method for charging for a response to a user query generated by a generative AI system. Additionally, it may be an objective of the present invention to establish an autonomous and self-assessing method that is user tailored for enabling resource allocation for a response, the method ensuring a seamless and secure user experience. The present invention solves the aforementioned problems by a method according to claim 1. Where the below description mentions price or pricing - this may in the embodiments claimed relate to energy usage. In particular, the problems are solved by a method for providing a response, the method consisting of: - receiving a user query (Q) from a user (10) by an integrated control system (20); - generating a request (PQ) based on the user query (Q); - transmitting the request (PQ) to a large language model (31), wherein the integrated control system (20) requests the large language model (31) to generate an expected energy usage indication; - receiving by the integrated control system (20) the expected energy usage indication generated by the large language model (31); - determining based on the expected energy usage a response resource; - requesting confirmation from the user (10) to allocate the response resource; - receiving by the integrated control system (20) a confirmation (C) to allocate the response resource; - allocating the response resource; - transmitting a response (A) associated with the user query (Q) to the user (10). The query stems from a user and may be issued through a client device, for instance a mobile device, personal computer or any other type of digital device. However, it is not necessary that client device is a dedicated hardware assigned to a single user. Indeed it is also possible that the client interacts with the same system that also hosts, at least partly the LLM and / or any other component as set out above. The user query is commonly a question or a prompt that inquires information or assistance regarding a specific topic and may be provided as text, an image, audio or video data, and / or a combination of these modalities. Correspondingly, the response to the user query may take the form of text, an image, audio / video data or a combination of these modalities. The received user query may be generated, for example, via web interface. Alternatively, the received user query may be generated by any other means, for example, via word processing software like Microsoft Word, a software running in a car, a third-party interface, application programming interface (API), messaging apps or any other type of device or interface. It is one aspect of the invention, to use generative AI, in particular an LLM to arrive at price indication for a user query which is also to be answered using an LLM. The LLM used for establishing an expected energy usage indication can be the same as the LLM used for processing the user query and providing the response. Alternatively, the LLM used for generating this indication can be a separate LLM or any other generative AI mechanism. This indication can be defined as the (estimated) expected energy usage for carrying out the query. This can comprise multiple factors to influencing the energy usage. In some embodiments this energy usage may have an effect on the price charged for carrying out the query. A price indication may also comprise a total estimated cost for a response. Alternatively, the price indication may be a combination of these approaches, in particular it can comprise the individual cost for each factor influencing the pricing and the total estimated cost for the response. Allocation of a price, in particular the response price, may refer to the process of setting aside, e.g. in a digital wallet, an amount or value from a user that is equal to the amount of the price indication. An allocation process does not necessarily initiate a payment transfer. For instance, an actual payment may be carried out only when the allocated amount crosses a certain threshold. One advantage of the allocation is the reduced number of individual transactions and transaction fees when charging for generated responses. One general advantage of the invention is that owing to the increased energy efficiency as a result of queries not going forward due to expected energy over a pre-set threshold and because of the expected energy usage driving users to sue the service at different times, and due to user feedback training the system to be more efficient, the overall performance of the service can be enhanced. For example, the savings due to energy efficiency can be used to cover expenses related to hardware components, balance the overall system load and / or provide a scalable infrastructure that can meet increased user demands and improve responsiveness. Additionally, copyrighted material sources can be paid for and used in generating responses, potentially improving the quality of the responses. Price indications for responses provided by the LLM are established using a pricing request that is sent to the LLM. In one embodiment, generating the pricing request may comprise: - using a model template and / or - using a natural language processing model and / or - using a feedback mechanism, the feedback mechanism comprising: - receiving user feedback associated with the response (A) and / or the response resource; - storing the user feedback in a historian database (24); - analyzing the user feedback, e.g. to identify most frequent user complaints; - adapting the request based on the analyzed user feedback. The model template aims to provide a clear and well-defined framework for generating requests. It is structured in a way that instructs or guides the LLM to generate a model for a response to a user query. In particular, the template may instruct the LLM to generate a model based on provided criteria. The said criteria may include response performance indicators and / or data items. The response performance indicators may refer to qualitative measures which are predominantly subjective and can include, but are not limited to, for example, usefulness, complexity, feasibility, clarity and relevance, scientific value, entertainment value, potential impact on the user, or any other qualitative measure based on which the quality of a response can be evaluated and an expected energy usage indication can be provided accordingly. In an embodiment, some of the data items may be autonomously collected by the system, for instance device information or location. The system may also be provided with access to third-party cookies associated with the user which can be used to collect behavioral data items. In one embodiment, some of the data items may be received by the system. For instance, the resource system may store data about previous user queries which can be analyzed to gain insight on user preferences. Furthermore, the system may collect data items from various digital content providers, particularly to collect data that is not user related. According to one aspect of the invention, the method can assign and dynamically adjust weights to response performance indicators based on at least one data item. The advantage is that the method can consider the relative importance of response performance indicators with respect to, for instance, the location of the user, education, etc. In one embodiment, the method can dynamically adjust weights assigned to the response performance indicators based on any combination of data items. Therefore, a more important response performance indicator will be assigned a larger weight compared to a less important indicator. Response performance indicators that may not be considered significant, for instance, response performance indicators that have a weight below a certain threshold, will not be included as a parameter in the request. For example, an expected energy usage indication provided by the LLM that is adjusted according to the location of the user, the time of day, peak times for queries, timing of renewable energy availability etc.. In another example, this may allow the establishment of more personalized pricing strategies, such as offering discounts to users with lower or no income, particularly if they modify their behavior to increase energy efficiency of the system - hence making the service more affordable. Alternatively, adapting the weights with respect to any combination of user profile data items allows for a finer tuning of the importance of the indicators. In one embodiment, the weighting of the response performance indicators is performed using a neural network. Additionally different users may require different explanations to queries based on their personal characteristics. It is often seen in science education that explanations of different depths is appropriate for children at different learning stages. For example, the explanation of the complexity of an animal cell differs greatly for an 11 year old to an undergraduate biologist. Similarly, the explanation of how electrons move in a circuit or semiconductor change between this age range. When a user requests a response to a query by taking the user's education level, age etc. into account the answer to the query itself can change. Therefore, the energy usage will also change accordingly. Therefore, taking these user data into account can make the expected energy usage more precise, whilst also increasing user satisfaction with the service by providing answer's to queries at the correct level for the user. Adapting the request based on data items may yield customized resource needs to the clients which can potentially decrease churn rates and attract new users. In addition, considering that some data items are dynamically changing, the method is capable of capturing fluctuations in user behavior and adapting the pricing request accordingly. Another advantage, for example, is that analysis of previous queries of a user can provide insights about the energy usage a user is willing to make for a particular service or product, thus allowing to adapt the pricing request in a way to understand the depth of answer that may required to the query, and therefore the expected energy usage that will result. Of course, the LLM can also be requested to take into consideration previous requests - the course of the communication - for determining an appropriate expected energy usage indication. The model template can be based on a framework that commonly includes placeholders. The placeholders can take any value or variable, for instance age, location or any other attribute, criteria or characteristics relevant or important to the user, and can be selected in a way to provide a more personalized response to a request. Providing a variable or a value to a placeholder may be performed automatically, for example using a programming function that inserts a variable or a value into a designated placeholder. In an alternative, the model template can be automatically adapted using a natural language processing (NLP) model, for instance, a pre-trained NLP algorithm such as GPT, BERT (Bidirectional Encoder Representations from Transformers) or any other pre-trained model. Alternatively, the NLP model and / or algorithm can be trained on specific data sets relevant for generating a pricing request or rely on any type of data or rules, for example linguistics or statistical rules. To modify the pricing model template, the NLP model may be utilized to fill in each placeholder with the most contextually appropriate parameter from a plurality of highly relevant parameters. To be able to improve the quality of the request, the method further comprises a feedback mechanism that uses client feedback to refine the pricing request. In one embodiment, the feedback mechanism can be incorporated as a rating system and users are invited to evaluate, on a predetermined scale, a single or various categories related to the quality of the response to their queries. As an example, the feedback mechanism may include criteria such as how useful or important the responses are according to the user or if the user finds the determined price indication appropriate for the response. Alternatively, the feedback mechanism can take the form of a survey containing open questions that enables users to provide a more detailed and / or personal feedback. In general, the feedback mechanism can be incorporated in any manner suitable for receiving feedback that can be used to improve the generation of the request. The feedback can be stored in a historian database and analyzed to identify most frequent complaints based on which the request should be modified. To provide a user with an expected energy usage for allocation, the method determines a response resource. The response resource represents the energy cost to be allocated for providing the response to a given query or input. In particular, determining the response resource can comprise at least one of the following: - calculating a total amount based on the indication - establishing a response range using the indication After determining the response resource, the method can request a confirmation from the user to allocate the response resource in order to provide the generated response. The credit allocated to a user may be stored in a digital wallet integrated, for example, in the resource system. In an embodiment, the credit may be used to pay (or offset) a response resource generated for a different response to a query. In an alternative, the credit may also be used to purchase assets from any asset provider that may recognize the credit as a valid payment currency - or may be thought of as a carbon credit. This feature offers a higher flexibility to users with respect to their payment preferences and serves as an additional and reliable source of revenue to a platform utilizing this method step. At least some of the above given problems are also solved by a computer readable medium. The computer readable medium stores instructions that when executed by at least one processor cause the at least one processor to implement one of the methods / embodiments as described above. The object of the present invention is further solved by a system for providing response to a user query. The system comprises an integrated control system which is adapted to: - receive a user query (Q) from a user (10); - generate a request based on the user query (Q); - transmit the request to a large language model (31) and request the large language model (31) to generate an expected energy usage indication; - receive the expected energy usage indication generated by the large language model (31); - determine based on the expected energy usage indication a response resource; - request confirmation from the user (10) to allocate the response resource; - transmit a response (A) associated with the user query (Q) to the user (10). Regarding the system, similar or identical benefits result as described in connection with the above methods. Further embodiments are discussed in the dependent claims. In the following, embodiments of the invention are described with respect to the figures, wherein Fig. 1 shows an a user, an integrated control system, a large language model server and a resource system that are connected through the internet, Fig. 2 shows an illustrative diagram of a method for providing a response to a user query in accordance with the present invention, Fig. 3 shows the components of the integrated control system, which is used to carry out the method for providing a response to a user query, Fig. 4 shows an exemplary request utilized to generate an expected energy usage indication for a response to a user query. In the following description, same reference signs may be used for same parts and (different) parts with the same effect. Fig. 1 shows a user 10, connected to an integrated control system 20 and a re-source / payment system 40, preferably for conducting micropayments or recording accepted energy usages, through the internet 1. The integrated control system 20 comprises a user interface that provides a seamless access for the user 10 to the integrated control system 20. The user 10 interacts with the integrated control system 20 through a client device, such as but not limited to a mobile, a PC or any other hardware device or component that is compatible with the integrated control system 20. The integrated control system 20 is also communicatively connected with a large language model server 30, hosting a large language model 31, and the payment system 40. The integrated control system 20 can serve as a gateway to the large language model server 30 and the resource system 40. In an alternative, the resource system 40 can also directly communicate with the user 10. The integrated control system 20 receives queries from the user 10, processes the queries and forwards the queries to the large language model 31 hosted on the large language model server 30. Accordingly, the integrated control system 20 receives responses from the large language model server 30, wherein the responses are generated by the large language model 31 and are related to the queries. The integrated control system 20 may require the user to register an account for access. As part of the registering process, the integrated control system 20 may request the user to provide personal information, wherein the personal information refers to a data item or data items, such as age, bank information, educational status, etc. The personal information is collected, for instance using a questionnaire. Additionally, the integrated control system 20 may be capable of gathering information autonomously, for example by access to third-party cookies, resolving IP and / or MAC addresses, analyzing the type of software the user is using (what operating system is installed on the client device of the user 10) or by any other means for collecting information autonomously. The integrated control system 20 may also implement advanced security features, for instance two-factor or biometric authentication for providing secure access to its functionalities. Alternatively, the user may be registered with the payment system 40 and the integrated control system 20 may retrieve credentials for the user and / or personal information from the resource system 40. Whether or not the user is registered, the integrated control system 20 assigns a unique identification number (ID) to the user. The ID number is then stored in a general database 25 (see figure 3) together with other data, e.g. user credentials and personal information. The ID number is also provided to a historian database 24 (see figure 3), wherein the historian database 24 serves as a detailed historical repository for user activities, in particular interactions with the integrated control system 20 and / or the payment system 40. The ID number can be used as a common identifier for retrieving information about the user from the historian database 24 and the general database 25. Fig. 2 shows an exemplary method according to the invention. In an initial step, the integrated control system 20 receives a user query Q from the user 10 in the form of text, audio and / or video by means of a user interface. The integrated control system 20 stores the user query Q in the historian database 24 of the integrated control system 20 and analyzes the user query Q to extract information from it. For instance, the integrated control system 20 may be capable of executing a natural language processing model, stored in a memory 22 (see figure 3), which performs semantic analyses on the user query Q for identifying personal information, preferences of a user or trends in the user's requests to the large language model 31. Additionally, the semantic analyses may be used to identify the underlying reason for or motivation behind the user query Q and adapt a request PQ for the user query Q according to the specific intention of the user. In one embodiment, the natural language processing model is provided as part of the integrated control system 20. In another embodiment, the natural language processing model is implemented in a separate component, wherein the separate component is communicatively connected with the integrated control system 20. In one embodiment, the integrated control system 20 sends a request to the large language model 31 which instructs the large language model 31 to perform (semantic) analysis on the user query Q and to return the processed information to the integrated control system 20. In each embodiment, the extracted information is stored in the historian database 24 for the purpose of constructing a more detailed set of data for the user. The integrated control system 20 then generates the pricing request PQ, typically in the form of text. Upon generating the request PQ, the integrated control system 20 transmits the request PQ to the large language model 31 (see Fig. 2). The large language model 31 generates / initializes a model, as instructed by the request PQ. In one embodiment, the request PQ comprises the user query Q and the large language model 31 generates the model using the information included in the user query Q. The information in the request PQ can - in addition to the user query Q - contain criteria based on which the large language model 31 generates the model. The criteria can include qualitative measures, for example, complexity, impact, personalization level, etc., and / or data related to the user such as age, nationality, location, etc. The large language model 31 is adapted to interpret the user query Q and what information it inquires and assesses the user query Q based on the criteria provided in the request PQ. For example, the request PQ can request / instruct the large language model 31 to generate a model for the user query Q 'what is the capital of the United Kingdom?''assessing the complexity and usefulness' (=per-formance indicator) and 'considering that the user is a student' (= background information = data item). In a next step, the integrated control system 20 receives the model - descriptive text describing the pricing model - and the expected energy usage indication generated by the large language model 31 for the user query Q, wherein the indication is particularly an energy or price P (referred to herein as a price), for example 10 Eurocents or 10 mW (it may be expressed in either manner). In one embodiment, the integrated control system 20 receives the price P in the form of individual prices for each criterion included in the model. Continuing with the example above, the model may establish the following price P for the user query Q 'what is the capital of the United Kingdom?': complexity: 1 Eurocent usefulness: 2 Eurocents The integrated control system 20 processes the received price P and determines a response resource RP. In one embodiment, the integrated control system 20 determines the response RP by calculating the total amount of the individual prices or it might just pass on the price as provided by the large language model 31, e.g. the 10 Eurocents as captioned above. After generating the response resource RP, the integrated control system 20 transmits the response resource RP to the user 10 and requests a confirmation C for allocating the response resource RP. For example, the integrated control system 20 may request the confirmation C and transmit the response resource RP through a modal window that overlays the existing content, the modal window containing the response resource RP and interactive buttons, such as 'accept' or 'reject'the allocation of the response resource RP. In a next step and in accordance with one embodiment of the invention, the integrated control system 20 receives the confirmation C from the user for allocating the response resource RP. If the user rejects the allocation, e.g. by clicking on the 'reject' button, the integrated control system 20 may display a message, acknowledging the rejection and inviting the user to provide a new query. If the user accepts the allocation, the integrated control system 20 asks the payment system 40 to allocate the response resource RP by transmitting a request for allocation RAII to the resource system 40. The request for allocation RAII can comprise the response resource RP and / or the ID of the user and is constructed in a way that prompts the payment system 40 to initiate the allocation. For example, the request for allocation RAII can be an HTTP request sent to a dedicated API endpoint of the resource system 40. The request for allocation RAII can be provided in any format accepted by the resource system 40 and meeting the requirements of the payment system 40.The resource system may be a payment system - where the energy usage is stored as a financial data point. After the integrated control system 20 transmits the request for allocation RAII to the payment system 40, the payment system 40 allocates All the response resource RP from the user. In one embodiment, the resource system 40 initiates a transaction from the user only when the allocated response resource RP exceeds a particular threshold. If the response price RP is below the threshold, the resource system 40 can store the response resource RP, for example in a digital wallet (a dedicated database) and adds the response resource RP to the response resource RP provided for a different user query Q. The transaction is initiated when the accumulated amount exceeds the threshold. After the resource system 40 allocates All the response resource RP from the user, the integrated control system 20 receives a confirmation for allocation CAII from the payment system 40. For example, the integrated control system 20 can receive a notification by the payment system 40 every time the resource system 40 performs an allocation, the notification including the ID of the user. When the integrated control system 20 receives the confirmation for allocation CAII, it transmits a request AQ to the large language model 31 to provide a response A to the user query Q. When the integrated control system 20 receives the response A from the large language model 31, the integrated control system 20 simply forwards the response A to the user 10 in one embodiment. Fig. 3 shows individual components of the integrated control system 20. In the described embodiment, the integrated control system 20 has a processor 21 configured to execute instructions stored in a memory 22, a communication interface 23 allowing the integrated control system 20 to communicatively interact with the user 10, the large language model 31, hosted on the large language model server 30, and the resource system 40. Moreover, the integrated control system 20 has a historian database 24 and a general database 25 for storing data from the client device of the user 10, the large language model 31 and the resource system 40. The memory 22 stores a set of instructions for implementing a process control algorithm 27. When executed by the processor 21, the process control algorithm 27 carries method as described in connection with Fig. 2. In one embodiment of the invention, the historian database 24 and the general database 25 are dedicated blocks of the integrated control system 20 which store data utilized in the generation of the request PQ. In one embodiment, the historian database 24 and the general database 25 can be cloud databases and one of the steps of the process control algorithm 27 is to transmit and receive data for a user to and from the cloud databases, for instance using APIs. In one embodiment, the historian database 24 receives and stores data also from the resource system 40, wherein the data is related to previous purchases made by the user. The data may be received from the payment system 40, for instance after every purchase made by the user. Fig. 4 shows an exemplary request PQ for generating a model by the large language model 31. In this example, the request PQ comprises the user query Q and guides the large language model 31 to generate a model based on the user query Q and qualitative measures, namely entertainment value, utilitarian value, complexity, personalization and impact to the user (in this application also referred to as performance indicators). The exemplary request PQ provided is the following: "You are tasked as a market researcher with a specialization in pricing digital content. Your goal is to develop a pricing model in cents for responses given by ChatGPT to user queries or prompts. Please analyze and suggest pricing based on the following categories: - Entertainment Value: How enjoyable is the content to consumers? - Utilitarian Value: What tangible benefit or usefulness does the answer provide? - Complexity: Does the answer deal with simple or complex topic? - Personalization: Is the answer tailored to individual consumer needs or questions? - Impact: Consider the overall potential effect of the answer on a consumer's life or decision-making process Task: For each answer that ChatGPT provides to a specific query or prompt, apply the pricing criteria mentioned above. Provide a detailed breakdown and justification for the pricing in cents for each category. When you develop the pricing model as requested, apply the pricing model to the query 'What is the climate like in Antarctica?'." In response to the exemplary request PQ, the large language model 31 generates a model and applies the model to the user query Q. The model is summarized below: "Entertainment Value: High Entertainment Value (e.g., creative, humorous, engaging): 15 cents Moderate Entertainment Value (e.g., informative and engaging): 10 cents Low Entertainment Value (e.g., purely informative): 5 cents Utilitarian Value: High Utilitarian Value (e.g., actionable advice, solutions): 20 cents Moderate Utilitarian Value (e.g., informative but not highly actionable): 10 cents Low Utilitarian Value (e.g., general information): 5 cents Complexity: High Complexity (e.g., detailed explanations on intricate topics): 15 cents Moderate Complexity (e.g., covering a broad topic with some depth): 10 cents Low Complexity (e.g., straightforward and simple information): 5 cents Personalization: High Personalization (e.g., tailored to user's specific needs): 20 cents Moderate Personalization (addresses the query but not highly tailored): 10 cents Low Personalization (e.g., generic information): 5 cents Impact: High Impact (e.g.potentially life-changing advice): 25 cents Moderate Impact (e.g., influencing decision making): 15 cents Low impact (e.g., general information with minimal influence): 5 cents Applying the Model to the Query 'What is the climate like in Antarctica?': Entertainment Value: Low (5 cents) - as the query is focused on factual information Utilitarian Value: Moderate (10 cents) - providing valuable information about the climate. Complexity: Moderate (10 cents) - requiring a degree of detail in explaining the climate. Personalization: Low (5 cents) - the answer is not highly tailored to individual needs. Impact: Low (5 cents) - general information with minimal influence." In one embodiment, the request PQ does not include the user query Q and the integrated control system 20 transmits the user query Q to the large language model 31 after transmitting the pricing request PQ. In one embodiment, the integrated control system 20 may only transmit a summary of the original user query Q, e.g. 'Assuming you would be asked about ... , please provide a pricing model taking into consideration In one embodiment, the integrated control system 20 hosts the large language model 31. In particular, the integrated control system 20 stores the large language model 31 in the memory 22. In one embodiment, the confirmation C for allocation provided by the user is received directly by the resource system 40. In turn, the resource system 40 performs the step of allocating All the response resource RP and confirms the allocation CAII to the integrated system 20. In one embodiment, the integrated control system 20 receives the response A to the user query Q together with receiving the model and the expected energy usage indication. For example, the request PQ may additionally include instructions for the large language model 31 to generate the response A for the user query Q. Consequently, the method step AQ requesting the response A to the user query Q and the following method step of receiving the response A are excluded. At this point, it should be noted that all of the parts described above are claimed to be relevant to the invention when considered alone and in any combination, especially of the details shown in the drawings. In an alternative embodiment rather than indicating an expected energy usage, a pricing indication may be indicated. This may have an equivalent effect to the indication of an expected energy usage. However, in some embodiments may allow for further control of the pricing based on other factors or data items that do not influence energy usage such as employment status of the user etc. The data items may refer to, in particular user data items, such as but not limited to age, gender, nationality, education, or any other data item that provides personal information about the user. The data items may also refer to, for example, data items extracted from or related to the user browsing and search history, social media, asset purchases, etc. The data items may also comprise, for instance, quantifiable data items, such as location, weather, time of day or any other quantifiable data item that may be relevant for generating a price for a response. In one embodiment, some of the data items, in particular user data items, may stem directly from the user. For example, the user may be requested by the platform to register a profile or account and provide personal information. In one embodiment, the method adopts also an ad-supported model and provides a request to the user to watch a video. If the user accepts to watch the video, the method can further provide several options to the user. In particular, the method comprises the following steps: - sending a request to the user to watch a video; - determining whether the user has watched the video, when the user has watched the video: - reducing the response price by a specified amount and allocating the reduced price; and / or - allocating a credit to the user. Clauses as per this embodiment are shown below: Clauses 1. A computer-implemented method for providing response, the method consisting of: - receiving a user query (Q) from a user (10) by an integrated control system (20); - generating a pricing request (PQ) based on the user query (Q) and / or parameters attached to the user (10); - transmitting the pricing request (PQ) to a large language model (31), wherein the integrated control system (20) requests the large language model (31) to generate a price indication, in particular a price (P); - receiving by the integrated control system (20) the price indication generated by the large language model (31); - determining based on the price indication a response price (RP); - requesting confirmation from the user (10) to allocate the response price (RP); - receiving by the integrated control system (20) a confirmation (C) to allocate the response price (RP); - allocating the response price (RP), preferably using a payment system (40); - transmitting a response (A) associated with the user query (Q) to the user (10). 2. The method of clause 1, wherein the request for confirmation to allocate the response price (RP) comprises the response price (RP). 3. The method of any one of the preceding clauses, wherein generating the pricing request (PQ) comprises: - using a pricing model template and / or - using a natural language processing model and / or - using a feedback mechanism, the feedback mechanism comprising: - receiving user feedback associated with the response (A) and / or the response price (RP); - storing the user feedback in a historian database (24); - analyzing the user feedback, e.g. to identify most frequent user complaints; - adapting the pricing request (PQ) based on the analyzed user feedback. 4. The method of any one of the preceding clauses, in particular according to clause 3, wherein the pricing model template comprises at least one placeholder, wherein the at least one placeholder is assigned at least one response performance indicator or at least one user profile data item. 5. The method according to any one of the preceding clause, wherein at least one response performance indicator is provided to the large language model (31), e.g. as part of the pricing request (PQ), wherein the pricing request (PQ) causes the large language model (31) to generate the price indication based on the at least one response performance indicator. 6. The method according to any one of the preceding clauses, wherein at least one user profile data item is provided to the large language model (31), e.g. as part of the pricing request (PQ), wherein the pricing request (PQ) causes the large language model (31) to generate the price indication based on the at least one user profile data item. 7. The method according to any one of the preceding clauses, the method comprising: assigning and / or (dynamically) adjusting a weight for at least one response performance indicator by the integrated control system (20) based on at least one user profile data item, and forwarding the weight to large language model (31) for use generating the price indication. 8. The method according to any one of the preceding clauses, in particular according to clause 7, wherein the assigning and / or (dynamically) adjusting of weights is performed using a neural network. 9. The method according to any one of the preceding clauses, wherein determining the response price (RP) comprises at least one of the following: - calculating a total amount based on the price indication, - establishing a response price range using the price indication. 10. The method according to any one of the preceding clauses, wherein allocating the response price (RP) includes: transmitting a request for allocation (RAII) by the integrated control system (20) to the payment system (40); receiving by the integrated control system (20) a confirmation for the allocation (CAII) from the payment system (40) and / or from the user (10). 11. The method according to one of the preceding claims, in particular according to clause 1, wherein allocating the response price (RP) includes: transmitting from the payment system (40) to the user (10) a request to allocate the response price (RP), in particular after receiving a / the request for allocation (RAII) from the integrated control system (20); receiving by the payment system (40) confirmation (C) from the user (10) to allocate the response price (RP); allocating (All) by the payment system (40) the response price (RP). 12. The method of any one of the preceding clauses, in particular according to clause 1, wherein the method comprises: - sending a request to the user (10) to watch a video, - determining whether the user (10) has watched the video, when the user has watched the video: - reducing the response price (RP) by a specified amount and allocating the reduced price; and / or - allocating a credit to the user (10). 13. The method according to any one of the preceding clauses, in particular according to clause 11, wherein the payment system (40) comprises a digital wallet for allocating the credit. 14. A computer readable medium (22) storing instructions that when executed by at least one processor (21) cause the at least one processor (21) to implement the method according to any one of the preceding clauses. 15. A system for providing response to a user query (Q), the system comprising an integrated control system (20), wherein the integrated control system (20) is adapted to: - receive a user query (Q) from a user (10); - generate a pricing request (PQ) based on the user query (Q) and / or parameters attached to the user (10); - transmit the pricing request (PQ) to a large language model (31) and request the large language model (31) to generate a price indication, in particular a price (P); receive the price indication generated by the large language model (31); determine based on the price indication a response price (RP); request confirmation from the user (10) to allocate the response price (RP); transmit a response (A) associated with the user query (Q) to the user (10). Reference signs: 10 user 20 integrated control system 21 processor 22 memory 23 communication interface 24 historian database 25 general database 30 LLM Server 31 large language model (LLM) 40 payment system Q query PQ pricing request P price RP response price C confirmation A response RAII request for allocation CAII confirmation for allocation All allocation

Claims

1. A computer-implemented method for providing response, the method consisting of:- receiving a user query (Q) from a user (10) by an integrated control system (20);- generating a request (PQ) based on the user query (Q);- transmitting the request (PQ) to a large language model (31), wherein the integrated control system (20) requests the large language model (31) to generate an expected energy usage indication;- receiving by the integrated control system (20) the expected energy usage indication generated by the large language model (31);- determining based on the expected energy usage a response resource;- requesting confirmation from the user (10) to allocate the response resource;- receiving by the integrated control system (20) a confirmation (C) to allocate the response resource;- allocating the response resource;- transmitting a response (A) associated with the user query (Q) to the user (10).

2. The method of claim 1, wherein the request for confirmation to allocate the response resource comprises the response resource.

3. The method of any one of the preceding claims, wherein generating the request comprises:- using a model templateand / or- using a natural language processing modeland / or- using a feedback mechanism, the feedback mechanism comprising:- receiving user feedback associated with the response (A) and / or the response resource;- storing the user feedback in a historian database (24);- analyzing the user feedback, e.g. to identify most frequent user complaints;- adapting the request based on the analyzed user feedback.

4. The method of any one of the preceding claims, in particular according to claim 3, wherein the model template comprises at least one placeholder, wherein the at least one placeholder is assigned at least one response performance indicator.

5. The method according to any one of the preceding claims, wherein at least one response performance indicator is provided to the large language model (31), e.g. as part of the request, wherein the request causes the large language model (31) to generate the expected energy usage indication based on the at least one response performance indicator.

6. The method according to any one of the preceding claims, wherein at least one user profile data item is provided to the large language model (31), e.g. as part of the request , wherein the request causes the large language model (31) to generate the expected energy usage indication based on the at least one user profile data item.

7. The method according to any one of the preceding claims, the methodcomprising:assigning and / or (dynamically) adjusting a weight for at least one response performance indicator by the integrated control system (20) based on at least one variable such as the time of the day and / or overall load on the system, andforwarding the weight to large language model (31) for use generating the expected energy usage indication.

8. The method according to any one of the preceding claims, in particular according to claim 7, wherein the assigning and / or (dynamically) adjusting of weights is performed using a neural network.

9. The method according to any one of the preceding claims, wherein determining the response resource comprises at least one of the following:- calculating a total amount based on the expected energy usage indication,- establishing a response resource range using the expected energy usage indication.

10. The method according to any one of the preceding claims, wherein allocating the response resource includes:transmitting a request for allocation (RAII) by the integrated control system (20) to the resource system (40);receiving by the integrated control system (20) a confirmation for the allocation (CAII) from the resource system (40) and / or from the user (10).

11. The method according to one of the preceding claims, in particular according to claim 1, wherein allocating the response resource includes:transmitting from the resource system (40) to the user (10) a request to allocate the response resource, in particular after receiving a / the request for allocation (RAII) from the integrated control system (20);receiving by the resource system (40) confirmation (C) from the user (10) to allocate the response resource;allocating (All) by the resource system (40) the response resource.

12. A computer readable medium (22) storing instructions that when executed by at least one processor (21) cause the at least one processor (21) to implement the method according to any one of the preceding claims.

13. A system for providing response to a user query (Q), the system comprising an integrated control system (20), wherein the integrated control system (20) is adapted to:- receive a user query (Q) from a user (10);- generate a request based on the user query (Q);transmit the request to a large language model (31) and request the large language model (31) to generate an expected energy usage indication;receive the expected energy usage indication generated by the large language model (31);determine based on the expected energy usage indication a response resource;request confirmation from the user (10) to allocate the response resource;transmit a response (A) associated with the user query (Q) to the user (10).Amendment to the claims have been filed as followsAmended claims1. A computer-implemented method for providing response, the method consisting of:- receiving a user query (Q) from a user (10) by an integrated control system (20);- generating a request (PQ) based on the user query (Q);- transmitting the request (PQ) to a large language model (31), wherein the integrated control system (20) requests the large language model (31) to generate an expected energy usage indication;- receiving by the integrated control system (20) the expected energy usage indication generated by the large language model (31);- determining based on the expected energy usage a response resource;- requesting confirmation from the user (10) to allocate the response resource;- receiving by the integrated control system (20) a confirmation (C) to allocate the response resource;- allocating the response resource;- transmitting a response (A) associated with the user query (Q) to the user (10),the method further comprising:- assigning and / or (dynamically) adjusting a weight for at least one response performance indicator by the integrated control system (20) based on at least one variable such as the time of the day and / or overall load on the system, andforwarding the weight to large language model (31) for use generating the expected energy usage indication,wherein the assigning and / or (dynamically) adjusting of weights is performed using a neural network.

2. The method of claim 1, wherein the request for confirmation to allocate the response resource comprises the response resource.

3. The method of any one of the preceding claims, wherein generating the request comprises:- using a model templateand / or- using a natural language processing modeland / or- using a feedback mechanism, the feedback mechanism comprising:- receiving user feedback associated with the response (A) and / or the response resource;- storing the user feedback in a historian database (24);- analyzing the user feedback, e.g. to identify most frequent user complaints;- adapting the request based on the analyzed user feedback.

4. The method of any one of the preceding claims, in particular according to claim 3, wherein the model template comprises at least one placeholder, wherein the at least one placeholder is assigned at least one response performance indicator.

5. The method according to any one of the preceding claims, wherein at least one response performance indicator is provided to the large language model (31), e.g. as part of the request, wherein the request causes the large language model (31) to generate the expected energy usage indication based on the at least one response performance indicator.

6. The method according to any one of the preceding claims, wherein at least one user profile data item is provided to the large language model (31), e.g. as part of the request , wherein the request causes the large language model(31) to generate the expected energy usage indication based on the at least one user profile data item.

7. The method according to any one of the preceding claims, wherein determining the response resource comprises at least one of the following:- calculating a total amount based on the expected energy usage indication,- establishing a response resource range using the expected energy usage indication.

8. The method according to any one of the preceding claims, wherein allocating the response resource includes:transmitting a request for allocation (RAII) by the integrated control system (20) to the resource system (40);receiving by the integrated control system (20) a confirmation for the allocation (CAII) from the resource system (40) and / or from the user (10).

9. The method according to one of the preceding claims, in particular according to claim 1, wherein allocating the response resource includes:transmitting from the resource system (40) to the user (10) a request to allocate the response resource, in particular after receiving a / the request for allocation (RAII) from the integrated control system (20);receiving by the resource system (40) confirmation (C) from the user (10) to allocate the response resource;allocating (All) by the resource system (40) the response resource.

10. A computer readable medium (22) storing instructions that when executed by at least one processor (21) cause the at least one processor (21) to implement the method according to any one of the preceding claims.

11. A system for providing response to a user query (Q), the system comprising an integrated control system (20), wherein the integrated control system (20) is adapted to:- receive a user query (Q) from a user (10);- generate a request based on the user query (Q);- transmit the request to a large language model (31) and request the large language model (31) to generate an expected energy usage indication;- receive the expected energy usage indication generated by the large language model (31);- determine based on the expected energy usage indication a response resource;- request confirmation from the user (10) to allocate the response resource;- transmit a response (A) associated with the user query (Q) to the user (10),- assign and / or (dynamically) adjust a weight for at least one response performance indicator based on at least one user profile data item, andforwarding the weight to large language model for use generating the price indication,wherein the assigning and / or (dynamically) adjusting of weights is performed using a neural network.

Images