AI task operating system with embedded thinking power evaluation double models

Abstract

The invention discloses an AI task operation system with embedded thinking power evaluation double models, which supports natural language and graph structure language input, drives an AI model to execute a reasoning path through Prompt + JSON + TP control languages, and realizes task information structuring and clear target expression. Comprising a TP scoring model and a TI scoring enhancement module which are respectively used for scoring the basic adaptability and the structure adjustment capability of a task path to form a dual-model task scheduling mechanism. An enhanced thinking score TP'is generated through score fusion, path judgment and scheduling are conducted through a task execution controller TE-Control, an intelligent task execution closed loop is constructed in combination with an output structure of a T-Level output control module and a feedback mechanism of a CSI cognitive score and feedback module, the adaptability and the output matching degree of complex tasks are improved, and the task execution efficiency is improved. And the consistency of the response of the AI to the user target is enhanced. The system has closed scoring and scheduling closed loop, and has self-evolution ability.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence application systems, and more specifically, to an intelligent task management operating system embedded with a dual model for evaluating thinking ability, namely an AI task management operating system (AIOS), which is used to enhance the cognitive reasoning ability, execution control ability, and dynamic adaptation ability of AI in complex task environments. Background Art

[0002] Existing AI systems mostly rely on cloud model training and task execution, and there are problems such as privacy risks and model instability. The present invention designs a trinity architecture to run the AIOS task control system locally, realizing rapid adaptation and independent training of open-source AI models.

[0003] With the rapid development of large artificial intelligence models in the fields of natural language processing, task reasoning, and decision support, traditional AI task management systems are facing a series of bottlenecks and challenges, which are mainly reflected in the following aspects: (1) Uncertainty and unstructured problems in task input: Most existing AI task management systems rely on users to describe task requirements in natural language. However, natural language has problems such as vague expression, implicit structure, and logical jump, resulting in easy understanding deviation when the AI system parses task intentions, which in turn affects the accuracy of reasoning path generation and execution control.

[0004] (2) Lack of standardization and dynamic adjustment ability in reasoning path generation: Existing systems often generate static reasoning paths based on single input, and it is difficult to dynamically optimize reasoning flow according to task execution feedback. They lack the ability to adjust reasoning paths, reallocate resources, and call modules based on execution status, resulting in insufficient flexibility of system execution and weak ability to adapt to complex environmental changes.

[0005] (3) Lack of internal quality control mechanism in the cognitive reasoning process: During the execution of traditional AI tasks, there is a lack of an internal quality evaluation system for cognitive reasoning, and it is impossible to evaluate reasoning depth, reasoning speed, and reasoning accuracy in real time, resulting in easy accumulation of reasoning deviation in multiple rounds of reasoning, ultimately affecting the reliability of output results and user experience.

[0006] (4) Lack of standardized structure input and reasoning optimization support tools: Currently, there is a lack of a unified tool or language system that can input task information in a structured and standardized manner and cooperate with natural language parsing to support reasoning path generation, dynamic adjustment, and feedback optimization.

[0007] In response to the above problems, the present invention proposes an AI task management operating system (AIOS) embedded with a dual model for evaluating thinking ability. Summary of the Invention

[0008] The object of the present invention is to propose an AI task operating system (AIOS) with an embedded dual model for evaluating thinking ability in view of the deficiencies of the prior art. The AIOS system of the present invention combines a task input mechanism of natural language and standardized logic diagram language, combines an internal NLP semantic parsing engine, a task scheduling engine module, and a thinking ability scheduling and feedback optimization module, and guides AI to activate its three core thinking dimensions - thinking speed (execution efficiency of task response), thinking depth (logical level of path construction), and thinking accuracy (matching degree between output result and user goal) through structured task language, and quantitatively evaluates the above dimensions through the TP thinking ability scoring mechanism, realizing an intelligent closed-loop control from task understanding, reasoning path generation, dynamic feedback adjustment to final task output. Among them, the logic diagram language input mechanism is used as an auxiliary tool to improve the structuring and standardization degree in the task input stage, provide clear task module relationships, execution paths and feedback control logics for the system, thereby optimizing the overall task reasoning efficiency and cognitive control ability of the AIOS system.

[0009] The object of the present invention is achieved by the following technical solutions: an AI task operating system with an embedded dual model for evaluating thinking ability, the system comprising: A task input module, configured to receive a natural language description and / or a logic diagram language structure input by a user, and generate standardized task input information; An NLP semantic parsing module, configured to perform semantic parsing on the natural language input, extract task objectives, logical relationships and execution conditions, and fuse and parse with the logic diagram structure to generate a unified standardized task input structure; A task scheduling engine module, configured to generate control language instructions according to the parsed task structure and construct an inference path; A thinking ability scheduling and feedback optimization module, configured to comprehensively evaluate the thinking depth TD, thinking speed TS and thinking accuracy TA based on the thinking ability evaluation model TP, and dynamically adjust the inference path and module call according to the scoring result; A structured output processing module, configured to generate output contents with different precisions and cognitive depths according to the cognitive output result after the inference is completed; A user behavior database module, configured to record the operation behaviors, feedback information and cognitive preference data of the user during the task execution process; A cognitive collaboration module, configured to collaborate with the underlying AI model to perform task reasoning and cognitive output through structured input, inference control and dynamic feedback optimization.

[0010] Further, the task input module further includes: A natural language input module, configured to receive the task description information expressed by the user in natural language, and extract the task objectives, logical relationships and execution conditions through the NLP semantic parsing module; A logic diagram language input module, which is used to receive task nodes, execution paths, control logics, and parameter conditions expressed in a standardized logic diagram form.

[0011] Further, the task scheduling engine module further includes: A control language generator TML, which is used to generate internal control language instructions according to a standardized task input structure, and the control language instructions define task node calls, execution sequences, conditional judgments, exception handling, and alternate path settings; An inference path construction module, which is used to construct an inference path according to the control language instructions, and the inference path includes a main inference path and an alternate inference path and assigns corresponding path weights.

[0012] Further, the thinking ability evaluation model TP in the thinking ability scheduling and feedback optimization module includes: A thinking depth TD score, which is used to evaluate the hierarchical depth of task node calls and the tightness of inference logic during the inference process; A thinking speed TS score, which is used to evaluate the time consumption of inference path execution and the inference response efficiency; A thinking accuracy TA score, which is used to evaluate the consistency between the inference output result and the task goal or user expectation; A dynamic weight adjustment mechanism, which is used to dynamically adjust the weight coefficients of each index of TD, TS, and TA according to the task type, execution stage, or user preference, so that the total thinking ability score TP_Score optimally matches the task execution requirements; Wherein, the total thinking ability score TP_Score is calculated according to the following formula: TP_Score = λ1×TD + λ2×TS + λ3×TA In the formula, λ1 + λ2 + λ3 = 1, which are the weight coefficients corresponding to the dynamically adjusted TD, TS, and TA; A dynamic adjustment mechanism, which is used to dynamically adjust the inference path flow during the task execution process according to the inference feedback result, the change of thinking ability score, or the exception trigger condition, including switching to an alternate path, reallocating inference resources, or adjusting the execution sequence.

[0013] Further, the thinking ability scheduling and feedback optimization module further includes: A thinking ability enhancement mechanism, which introduces a thinking increment ΔTI based on a scoring enhancement module TI in the thinking ability enhancement mechanism, and incorporates the thinking increment into the total thinking ability score to form an enhanced thinking ability score TP′, where: When insufficient inference depth, decreased inference accuracy, or the inference path getting stuck is detected during the inference process, the TI module is automatically called; The TI module generates new auxiliary reasoning paths, alternative reasoning schemes, or reasoning branches, and sets the corresponding thinking increment ΔTI as supplementary cognitive value to be included in the adjustment of the thinking ability score. The calculation formula for the enhanced thinking ability score TP′ is: TP′ = TP + ΔTI Among them, TP represents the original total thinking ability score TP_Score, and ΔTI, as an enhanced scoring factor, combines the system task complexity and the historical task scoring trend, and participates in the dynamic adjustment of the weight coefficients of each index of TD, TS, and TA and the path optimization in real time during the task structure recognition or path reasoning process, constituting the dynamic enhancement mechanism of the system thinking ability score TP′.

[0014] Furthermore, the structured output processing module further includes a T-Level output control module. The T-Level output control module dynamically determines the cognitive depth and precision level of the output content according to the thinking ability score TP′ result, task complexity, and user requirements. The T-Level output control module includes: T1 level: Perform a quick response output based on the preliminary reasoning result, and the output content is concise with priority given to the response speed; T2 level: Add logical details and conditional analysis on the basis of T1 to provide a standard reasoning output; T3 level: Combine multi-angle reasoning and auxiliary path analysis to generate a deep reasoning output; T4 level: Introduce the new reasoning content generated by the scoring enhancement module TI to form a multi-dimensional comprehensive reasoning output; T5 level: Based on historical data, feedback evolution, and reasoning results, output a comprehensive analysis plan that combines deep reasoning and multi-dimensions.

[0015] Among them, the T-Level output control module dynamically upgrades or downgrades the output level according to the TP′ score during the task execution process to adaptively optimize the task output quality and response speed.

[0016] Furthermore, the user behavior database module is used to record the input content, execution feedback, operation behavior trajectory, and cognitive preference information of the user during the task execution process, extract the personalized preference characteristics of the user for reasoning depth, reasoning speed, and output granularity, and use this preference characteristic for task output control and execution path optimization.

[0017] Furthermore, the cognitive collaboration module is used to collaborate with the AI model to perform task reasoning and cognitive output, specifically including: A structured reasoning control mechanism for transmitting the standardized task input, reasoning path, and dynamic feedback instructions to the AI model through the control language TML to guide the AI model to perform the reasoning task; A dynamic thinking ability guidance mechanism, which is used to adjust the inference path, switch to an alternative path, activate the scoring enhancement module according to the real-time TP' score or TP_Score and the inference feedback, and transmit the adjustment instruction to the AI model; An innovative inference collaboration mechanism, which is used to call the innovative inference ability of the AI model during the inference process to generate new inference branches or auxiliary solutions; A cognitive feedback optimization mechanism, which is used to dynamically optimize the inference strategy according to the inference execution result and the user feedback, and continuously improve the overall cognitive inference effect.

[0018] Furthermore, the control language generator TML includes a logic diagram language control module, specifically as follows: (1) The graph language task node structure TML-Node, which represents the structural actions or judgment units of the task; (2) The graph language path connection structure TML-Path, which represents the logical jump and execution order between task steps; (3) The graph language control symbol structure TML-Control, which is used to express conditional judgment, exception handling, and path switching; (4) The graph language parameter label structure TML-Tag, which binds task parameters, user expectations, and output targets.

[0019] Furthermore, the AI task operating system, together with the personal PC and the open-source AI model platform, constitutes a three-in-one intelligent execution architecture; The AI task operating system is responsible for task input parsing, thinking ability scheduling, control path generation, and feedback closed-loop control; The personal PC has local computing power resources, a structured task execution memory space, and a task management and scheduling unit; The open-source AI model platform conducts two-way communication and task takeover with the AI task operating system through an interface protocol.

[0020] The beneficial effects of the present invention: 1. The present invention proposes an AI task operating system (AIOS) embedded with a dual-model thinking ability evaluation, which supports natural language and logic diagram structure task inputs, is compatible with the Prompt + JSON + control language hybrid mechanism, and combines a natural language parsing engine, an inference path generation mechanism, a thinking ability dynamic scheduling, and a feedback optimization mechanism to establish a complete cognitive closed-loop from task input to inference execution, feedback adjustment, thinking ability evaluation, and then to task output, significantly improving the task understanding ability, inference control ability, and execution adaptability of the AI system in complex environments. Among them, the logic diagram language system, as an auxiliary module, makes up for the deficiencies of natural language expression in terms of structure and standardization, and effectively supports the task parsing and inference dynamic control processes of the AIOS system by clearly defining task nodes, execution paths, control logics, and parameter labels.

[0021] 2. The present invention follows the design principle of "minimal invasion and maximum activation". It does not interfere with the underlying model structure and does not reconstruct the user interface. It maximally guides the large AI model to autonomously call its native thinking power resources to achieve intelligent and controllable human-machine task collaboration. It is adapted to mainstream large model systems including DeepSeek. The core of the invention lies in constructing a mechanism for the invocation, evaluation, and optimization of the "thinking power resources" inside the AI, making AIOS a transferable framework for guiding general models into domain-specific intelligent agents, and providing a new intelligent operation infrastructure of "thinking power as a service" for enterprises and institutions. The present invention is applicable to various application scenarios that require complex task management, cognitive reasoning optimization, and dynamic execution adjustment, such as intelligent manufacturing, smart healthcare, education and training systems, financial intelligent analysis and decision-making, etc.

[0022] 3. The present invention adopts a dual-model system composed of TP (Thinking Power Scoring Model) and TI (Thinking Power Scoring Enhancement Model). Through the TP score, it guides the basic path scheduling. When the score is insufficient or there are structural bottlenecks, the TI module is dynamically activated to generate an increment ΔTI. Based on the structural path performance and feedback behavior, an enhanced score TP′ is generated. The TE-Control module performs path judgment and module scheduling. At the same time, it cooperates with the T-Level output control module and the cognitive feedback channel of CSI to construct an intelligent closed-loop for task execution, providing a path basis for complex task structure optimization and continuous enhancement of model capabilities. This mechanism improves the flexibility of task path generation and the result matching degree, constituting the intelligent core of the thinking power control and execution scheduling of this system.

[0023] Therefore, the "enhancement" in the TI module belongs to a path regeneration mechanism triggered by the structural scheduling mechanism and can be recognized and executed by the controllable language logic. It has controllability, interpretability, and reproducibility, constituting the technical core for achieving structural optimization and improvement of complex task adaptation capabilities in the system of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is a schematic diagram of the principle architecture of an AI task operating system with an embedded dual-model for thinking power evaluation.

[0026] Figure 2 It is a schematic diagram of the TP scoring mechanism and path generation.

[0027] Figure 3 It is a schematic diagram of the basic structure composition of the logic diagram language system.

[0028] Figure 4 It is a schematic diagram of the standard process of the logic diagram language input AI system.

[0029] Figure 5 It is a schematic diagram of the conversion mechanism between the logic diagram language and the hybrid coding language.

[0030] Figure 6 It is a structural control diagram of the TP' enhanced scoring and TE path scheduling.

[0031] Figure 7 It is a logic diagram of the output quality evaluation and intelligent callback system.

[0032] Figure 8 It is a logic diagram of the structured task output and user configuration mechanism.

[0033] Figure 9 It is a logic diagram of the local AI cognitive execution control system.

[0034] Figure 10 It is a logic diagram of the local adaptive training driven by the user's business behavior. Specific implementation manners

[0035] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described below with reference to the accompanying drawings and implementation cases. It should be understood that the specific implementation cases described herein are only used to explain the present invention and are not used to limit the present invention.

[0036] As Figure 1 shown, it shows the full-process control logic of the system in key links such as user task input, semantic parsing, structure scoring, path scheduling, AI model call and cognitive feedback. In the task input stage of the AI task operating system AIOS with a dual model for evaluating thinking ability embedded in the present invention, in addition to the natural language input method, the system also supports task construction and execution process expression in the form of a structured logic diagram language. The user can use the logic diagram language to construct a task diagram, including task target nodes, judgment condition nodes, operation execution nodes, and the path connections and control symbols between them, and attach personalized task parameters such as execution levels and output methods through parameter tags. The system of the present invention uses a dual thinking ability evaluation mechanism composed of a TP scoring module and a TI scoring enhancement module (the second thinking ability evaluation model) as the central scoring engine, and combines a task execution controller TE-Control module, a Prompt generation module and a cognitive feedback module to realize intelligent selection of task execution paths, structure control and output optimization. Figure 1It reflects the complete closed loop of the system's "input - scoring - control - execution - feedback". As the basic scoring component for reasoning and scheduling, the refined structure of the dual - model scoring mechanism is shown in Figure 2 . Figure 2 This shows the task scoring and path control mechanism of the system based on the dual - thinking - ability evaluation model. After the user input is processed by the NLP semantic parsing engine, an initial execution strategy is generated and scored by the first thinking - ability model TP. The scoring result is initially scheduled by the TE - Control module according to the "in - path judgment logic". If the task complexity or goal is not clear, the second thinking - ability model TI will be activated for enhanced scoring to form the TP′ result, guiding the Prompt module to generate more accurate task instructions, thus realizing the closed - loop execution of path control and feedback scheduling. The TE - Control module contains two control algorithms: (1) The multi - round dynamic adjustment algorithm, mainly used for path scoring and feedback scheduling during task execution; (2) The adaptive algorithm, used for structure recognition and path initialization in the task input stage. The two cooperate to form the task execution controller (TE - Control), realizing the closed - loop control of task pre - understanding and mid - term scheduling.

[0037] The specific implementation of the present invention is as follows: After the system of the present invention receives the logical graph language structure, the logical graph language parsing engine identifies and analyzes the node structure and path logic. First, the parsing module extracts the types of each node and their connection relationships in the graph, identifies control symbols (such as sequence, branch, loop) to form a logical execution sequence; subsequently, the parameter tags are processed by the parameter translator into a format recognizable by the system (such as JSON fragments), and are sent to the hybrid - coding language generator together with the structure path.

[0038] The control language generator generates Prompt fragments according to the graph structure content, and generates a composite structure language of Prompt + JSON after fusing the parameter fields. This structured control language is sent to the AI large - model for reasoning and execution through the Prompt generation module.

[0039] The core structure of the control language generator (TML, Task Management Language) of the system of the present invention includes: Structure parsing unit: Accepts the input of the logical graph language structure, and identifies task nodes, paths, and parameters; Prompt template selector: Dynamically selects Prompt templates according to the task type, target output level (T - Level), and TP scoring result; Parameter field fusion module: Inserts user - input parameters, default task constraints, and graph structure path information into the Prompt template; Structured Output Generator: Outputs a composite control language structure in the form of "Prompt Instruction + JSON Control Parameters"; the generated result is sent by the Prompt generation module to the AI model interface for performing structured inference tasks.

[0040] The Logic Diagram Language Control Module is as follows: (1) The Diagram Language Task Node Structure TML-Node represents the structural actions or judgment units of the task; (2) The Diagram Language Path Connection Structure TML-Path represents the logical jumps and execution sequences between task steps; (3) The Diagram Language Control Symbol Structure TML-Control is used to express conditional judgments, exception handling, and path switching; (4) The Diagram Language Parameter Label Structure TML-Tag binds task parameters, user expectations, and output targets.

[0041] During the system operation, the CSI Cognitive Scoring and Feedback Module scores the AI output results, and the scoring results are fed back to the Logic Diagram Language Module to update the mapping rules and parameter translation logic, realizing the dynamic optimization of the graph structure execution strategy.

[0042] Through this implementation method, the Logic Diagram Language Module and the TP scoring mechanism, TE-Control Module, and Prompt generation module in the AIOS backbone system form a closed-loop execution path, realizing the whole-process structured task control logic from task input to execution feedback.

[0043] The logic diagram language mechanism in the present invention, as an expression method in the task execution manual stage, is mainly used to help users structurally express task objectives, execution modules, and control paths. It is a fast language tool for the AI task management system. Through the logic diagram language + natural language, complex task logical relationships can be quickly and accurately converted into a structured control language (TML) recognizable by the system, enabling the AI system to quickly understand and precisely call thinking resource capabilities to generate comprehensive solutions that meet user expectations.

[0044] The core innovation of the present invention lies in: dynamically evaluating the task complexity, reasoning depth, and output quality through the TP thinking ability scoring mechanism, and constructing a thinking path and scheduling process by the TE-Control Module to realize the intelligent control and quality feedback closed-loop in the task execution process.

[0045] Compared with the existing technical solutions centered on model structure control, graph learning inference, or model scrambling protection, the present invention has an essential difference in design principles and execution logics.

[0046] Example 1: Description of the Overall Architecture of the AIOS System: This embodiment provides an overall architecture of an AI task management operating system (AIOS) with an embedded dual-model for thinking ability evaluation, aiming to achieve an integrated intelligent closed-loop control of task input parsing, inference path generation, cognitive inference optimization, dynamic feedback adjustment, and thinking ability evaluation. The AIOS system mainly includes the following core modules, as Figure 1 shown: (1) Task input module: Supports users to describe task requirements through natural language, and at the same time introduces a standardized logical diagram language input method. The logical diagram language defines task nodes, execution paths, control logic, and parameter tags, assisting natural language input to form a structured and standardized task input system.

[0047] (2) NLP semantic parsing module: Performs semantic parsing on natural language input, extracts key elements such as task objectives, logical relationships, and execution conditions, and forms a fused parsing with logical diagram language input to generate a preliminary structured task description.

[0048] (3) Task scheduling engine module: According to the parsed task structure, generates internal executable control instructions through a control language generator (TML, Task Management Language) to construct an inference path. The inference path includes a main path, an alternative path, and weight settings, supporting dynamic feedback adjustment.

[0049] (4) Thinking ability scheduling and feedback optimization module: Built-in TP thinking ability evaluation model, which monitors in real time the inference depth (TD) related to the logical level of path construction, the inference speed (TS) related to the execution efficiency of task response, and the inference accuracy (TA) related to the matching degree between the output result and the user's goal during the inference process. Dynamically adjusts the inference path through TP scoring, and calls the scoring enhancement module (TI module) to increase the inference depth or new inference paths when necessary.

[0050] (5) Structured output processing module: According to the cognitive output result after inference, combined with the cognitive output level mechanism of the T-Level output control module, generates output content with corresponding accuracy and cognitive depth according to the task complexity and user requirement level.

[0051] (6) User behavior database module (UBB module): Records the operation behaviors and feedback correction information of users during task execution, serving as an important data source for the system's adaptive cognitive evolution.

[0052] (7) Cognitive collaboration mechanism module (AIOS + AI > 2): The "Cognitive Collaboration Module" refers to a module system in the system that is composed of a task structure expression (graph language module), an inference path scheduling (TP module), a cognitive output generation (T-Level), a cognitive feedback evaluation (CSI cognitive scoring and feedback module), and a parameter adjustment mechanism in coordination, and is used to support the AI model to perform the whole process of structural understanding, inference execution, and output feedback of cognitive tasks.

[0053] Through the cooperation of the above modules, the AIOS system can not only passively respond to user instructions, but also actively call the thinking power resources of the AI model based on the task structure, realize the improvement of cognitive intelligence, and the overall cognitive ability exceeds the level of a single AI model.

[0054] The overall architecture design forms the following closed-loop process:

Task Input

Semantic Parsing

Inference Path Construction

Dynamic Scheduling of Thinking Power

Cognitive Output Generation

User Feedback Collection and Adaptive Evolution

[0055] Example 2: Implementation of task input and structured parsing mechanism: This embodiment details the specific implementation method of the task input and structured parsing mechanism in the AIOS system. To solve the problems of fuzzy description, logical jump, and parsing difficulty in the task input stage of traditional AI systems, the present invention adopts a dual-channel task input mode that combines a natural language input module and a standardized logical graph language input module, effectively improving the structural degree of task description and parsing accuracy. See Figure 3 , which shows the core structural system of the logical graph language, including four basic units in task expression: task node unit (M1), path connection unit (M2), control symbol unit (M3), and parameter label unit (M4). Figure 3 Block A in defines the structural composition module of the logical graph language, clearly presenting the source and classification principles of each unit. Figure 3 Block B in provides a typical example of logical graph language expression, showing the path flow structure of the task process from data analysis, judgment conditions to execution operations. The output level L2 is marked by parameter labels, indicating the output control level corresponding to this structural statement. Figure 3 As the structural basis of the AIOS logical graph language module, core mechanisms such as the generation of the inference path planning and control language (TML) constitute the underlying expression specification of the logical graph language system in the system of the present invention.

[0056] (1) Natural language task input The user submits task requirements through natural language, including but not limited to: 1. Task Goal 2. Execution Conditions 3. Input Parameters 4. Output Requirements 5. Constraints The internal NLP semantic parsing engine of the system performs the following processing on the natural language text: Keyword extraction: Identify key task words and concepts.

[0057] Logical relationship identification: Analyze the causal, sequential, and dependency relationships between tasks.

[0058] Structure extraction: Abstract the task description into a parsable data structure (such as a task tree or task graph).

[0059] (2) Logic graph language task input To make up for the deficiency of natural language in structured expression, the system also supports logic graph language input, specifically including: Node Definition: Define each subtask or operation unit clearly.

[0060] Execution Path: Describe the logical flow relationship between task nodes, including sequential execution, parallel execution, conditional jump, etc.

[0061] Control Logic: Set trigger conditions, judgment conditions, execution constraints, etc.

[0062] Parameter Tags: Mark the parameter information required or output by task nodes.

[0063] The logic graph language is input into the system in a structured data form (such as JSON or graph structure model) and integrated with the natural language parsing results. For details, see Figure 4, which shows the standard conversion process of the logic graph language from structural input to hybrid language generation. The system constructs the task structure uniformly through three types of input elements: "graph language structure + parameter tags + control logic", via a structure builder, a symbol generator, and a parameter embedder. The constructed graph language structure will enter processing modules such as a mapping rule library and a parser to generate Prompt fragments and JSON control parameters, which are fused to form a structured control language, that is, the hybrid language format of Prompt + JSON. Finally, the control language is sent to the local AI model execution module through the Prompt + JSON output module to trigger the task inference process. The inference result is fed back to the CSI cognitive scoring and feedback module for rule correction and parameter optimization, forming a closed-loop path of "structure generation → inference execution → output evaluation".

[0064] (3) Fusion parsing mechanism The system adopts a fusion parsing mechanism to align and integrate the natural language parsing structure and the logic graph language structure to form a unified initial inference structure. The specific steps are as follows: 1. Modularly split the natural language parsing result and map it to the four elements of the graph language structure: Node, Path, Control Logic, and Parameter Tag, as the basic input of the structured control language; 2. Map the parsed modules to the logic graph nodes and paths; 3. Perform supplementary parsing and automatic inference on the ambiguities existing in the mapping process; 4. Generate a standardized task structure body as the basis for subsequent inference path generation and execution control.

[0065] (4) Standard input example For example, the user submits a natural language description: "Analyze customer data, classify customers according to age and purchase frequency, and predict future purchase potential." At the same time, the submitted node and path relationships defined by the logic graph language are as follows: Node 1: Data collection → Node 2: Customer classification (based on age) → Node 3: Customer classification (based on purchase frequency) → Node 4: Purchase potential prediction.

[0066] The modular split results are as follows: Node (Node): Customer data, age, purchase frequency, purchase potential Path (Path): Starting from "customer data", analyze "age" and "purchase frequency" in sequence to form a classification path, and finally point to "predict purchase potential" Control Logic (Control): If age < 30 and purchase frequency is high → high-potential customers; otherwise, they are general customers Parameter Tag: Classification Threshold = Age 30, Frequency Threshold = 5 times per month After system integration and analysis, a standardized initial inference structure is formed for subsequent processing by the control language generator (TML).

[0067] Example 3: Control Language Generation and Inference Path Construction: This example details the specific implementation of the control language (TML) generation and inference path construction mechanism in the AIOS system.

[0068] To achieve efficient conversion of task parsing results into AI-internal executable instructions, the present invention introduces a control language generator (TML: Task Management Language), and through an inference path construction mechanism, ensures that the task execution process has high flexibility, high adaptability, and dynamic adjustment capabilities. See Figure 5 , which shows the complete mechanism path for converting the logical graph language structure into an AI-executable hybrid language format (Prompt + JSON). After the user inputs the logical graph language content, the system first completes the construction of the structured task graph through the node builder, control symbol generator, and parameter tag embedder, and then enters the graph structure parser, mapping rule library, and parameter translator to generate the structural semantics, prompt fragments, and JSON control fields corresponding to the logical graph. The generated Prompt + JSON structure is transmitted to the local AI model (such as DeepSeek) through the Prompt + JSON output module for inference execution. The inference result is fed back to the CSI cognitive scoring and feedback module, which is further used for rule optimization and parameter correction to construct a two-way closed-loop mechanism of "graph structure → hybrid language → model inference → CSI feedback".

[0069] (1) The specific implementation of control language generation (TML generator) is as follows: The core functions of the control language generator include: Instruction Structuring: Convert the integrated and analyzed task structure body into a standardized control instruction set, and the instructions include basic operation units such as task invocation, conditional judgment, loop execution, and exception handling.

[0070] Semantic Fidelity: Ensure that the control language strictly retains the original semantics and execution intentions of the natural language and logical graph language inputs during the conversion process.

[0071] Extensibility Design: The control language supports module expansion to facilitate the subsequent addition of new task module and inference module instructions.

[0072] An example of the control language format is as follows (simplified version): json "TaskID": "T001", "TaskName": "Customer Data Classification", "ExecutionFlow": {"Node": "Data Collection", "Action": "Collect", "Parameters": {"DataSource": "CRM Database"}}, {"Node": "Age Classification", "Action": "Classify", "Criteria": "Age"}, {"Node": "Purchase Frequency Classification", "Action": "Classify", "Criteria": "PurchaseFrequency"}, {"Node": "Purchase Potential Prediction", "Action": "Predict", "Model": "Potential Prediction Model"} } After the control language instruction is formed, it enters the inference path construction stage.

[0073] (2)The specific implementation of the inference path construction mechanism is as follows: The main task of the inference path construction mechanism is to generate an inference path diagram that conforms to the task logic flow and execution conditions based on the control language instruction set. The construction process includes: Main path generation: Generate the basic inference main path (Main Inference Path) according to the task logic sequence.

[0074] Backup path design: Configure backup paths (Backup Path) for key task nodes to automatically switch and execute in case of inference failure, exception, or feedback deviation, ensuring high robustness of the system.

[0075] Path weight assignment: Assign weights (Weighting) to multiple possible paths, and the system dynamically selects the optimal path based on the thinking ability score during the inference process.

[0076] Dynamic Adjustment Point Setting: Set dynamic adjustment points (Adjustment Points) in the inference path, allowing the system to adjust the inference flow in real time according to feedback information during execution. The "dynamic adjustment points (AdjustmentPoints)" in this system refer to the preset structural node positions in the logical graph language path structure. This node is used to perform conditional jumps or parameter adjustment operations on the execution flow after receiving feedback inputs (such as TP scores, CSI results, user behavior preferences) during inference execution. Adjustment points are usually set at key nodes or parameter branch nodes in the path, and their functions include: 1. When the scoring result deviates from the preset threshold, interrupt the original path and switch to an alternative branch path (such as the TP' path); 2. Adjust the T-Level output level structure according to the CSI feedback and reconstruct the output method; 3. When the task structure does not fully match the user's goal, inject UBB business behavior characteristics and reconstruct the path parameters.

[0077] The inference path is managed internally in the system in the form of a directed graph. Each node (task unit) and edge (execution relationship) carry conditional tags and scoring metrics.

[0078] (3)Inference Path Example (Simplified Version) [Data Collection] → [Age Classification] → [Purchase Frequency Classification] → [Purchase Potential Prediction] ↘ (Alternative) [User Interest Classification] The main path sequentially executes data collection → classification → prediction; If an abnormality occurs at the purchase frequency classification node, switch to the alternative path to execute the user interest classification module.

[0079] (4)Linkage between Inference Path and Dynamic Feedback The inference path is not static. The system will adjust it in real time according to dynamic feedback (such as changes in TP thinking ability scores, task execution deviations) during execution: Specifically, it includes: adjusting the node order, dynamically invoking alternative modules, reconfiguring parameters and conditions, and increasing or decreasing the depth or breadth of inference; This mechanism ensures that the AIOS system can still maintain the flexibility and efficiency of inference in a complex environment.

[0080] Example 4: Thinking Ability Evaluation Model (TP Model) and Inference Dynamic Guidance: This example details the thinking ability evaluation model (TP model) embedded in the AIOS system and its application mechanism in inference dynamic guidance.

[0081] To achieve real-time cognitive evaluation and inference quality control during the inference process, the present invention designs a thinking power evaluation system (TP model), and dynamically guides the inference path and module scheduling based on the scoring results, improving the overall cognitive inference ability and execution adaptability of the system.

[0082] (1)Design of the TP Thinking Power Evaluation Model The TP (Thinking Power) model is the core cognitive evaluation engine of the AIOS system and consists of the following three basic indicators: TD (Thought Depth) measures the hierarchical depth and logical rigor of node calls in the inference path.

[0083] TS (Thought Speed) measures the time or computing resource consumption required to complete a full inference path, reflecting the inference efficiency.

[0084] TA (Thought Accuracy) measures the consistency between the inference output result and the task goal and user expectation, reflecting the inference quality.

[0085] Each indicator is scored separately, and different weights are assigned according to the task nature, and a comprehensive TP overall evaluation (TP_Score) is formed.

[0086] (2)TP Comprehensive Scoring Formula The formula for calculating the total TP score is as follows: TP_Score = λ1×TD + λ2×TS + λ3×TA Where: λ1, λ2, λ3 are dynamically adjustable weight factors, satisfying: λ1 + λ2 + λ3 = 1 The weight factors can vary dynamically according to task requirements, execution stages, and user preferences.

[0087] For example, in the "customer behavior prediction task", if the task is in the initial data modeling stage, the system will pay more attention to the inference depth (TD), and the weight factors can be set as λ1 = 0.6, λ2 = 0.3, λ3 = 0.1; When the system enters the real-time response scenario (such as an online recommendation system), the system pays more attention to the inference speed (TS), and the weights can be dynamically adjusted to λ1 = 0.2, λ2 = 0.6, λ3 = 0.2; If the user clearly indicates the need for a readable output, the proportion of output accuracy (TA) increases, and it is adjusted to λ1 = 0.3, λ2 = 0.2, λ3 = 0.5.

[0088] The above weight factors change dynamically according to the task stage, preference records in the user behavior database, and output feedback metrics, thereby enhancing the fitting and interpretability of the TP score for task performance.

[0089] The weight factor of TP is fine-tuned through the CSI cognitive scoring and feedback module: CSI = α×PMR + β×(1 - MIS) + γ×OHR Among them, α, β, and γ are dynamic weights that can be adjusted according to the task type. PMR (Prompt Matching Rate): The degree of matching between the output content and the user's input task goal; MIS (Missing Information Score): The degree of missing key information in the output; OHR (Output Heuristic Richness): Evaluation of the heuristic, depth, and diversity of the output content; The system internally sets a 3W module (What, Why, When) to parse the task goal, motivation background, and time requirements in the user input. This module is embedded in the task input stage as a preprocessing mechanism to enhance the context adaptability of TP evaluation and Prompt generation, and ensure accurate semantic matching of tasks.

[0090] The output result of the 3W module will be written into the input preprocessing buffer of the TP model and used as one of the reference factors for the integrity of the task semantic structure in the scoring calculation. Thus, the detection and enhancement of semantic clarity and intention integrity are completed prior to TD, TS, and TA scoring, constituting one of the important preconditions for the task scoring mechanism of the TP model.

[0091] (3) Inference dynamic guidance mechanism The scoring result of the TP model is directly applied to inference execution control, and the specific guidance methods are as follows: Dynamic adjustment of the inference path When the TP_Score drops below the preset threshold, the system automatically triggers the inference path adjustment mechanism, including: switching to an alternative inference path; adjusting the order of inference nodes or adding auxiliary nodes; activating the scoring enhancement module (TI module) to introduce new inference branches.

[0092] Dynamic scheduling of module tasks According to the TP scores at different inference stages, dynamically select the most suitable module or model resources for the current state to optimize inference efficiency and output quality.

[0093] Adaptive adjustment of parameters Automatically adjust inference parameters (such as node weights, path priorities, resource allocation ratios) to enhance the cognitive adaptability of the inference process.

[0094] Enhanced thinking trigger When the reasoning reaches a bottleneck (such as continuous decline in multiple scores), the system activates the TI module according to the TP score, calls the innovative reasoning ability of the AI model, and generates new reasoning directions or supplementary solutions.

[0095] (4) Dynamic guidance example process 1. The initial reasoning path is executed according to the standard task structure; 2. During the execution process, monitor TD, TS, and TA indicators in real time; 3. When the TS (speed) score drops below the threshold, it indicates insufficient reasoning efficiency; 4. The system adjusts the reasoning path, skips low-priority nodes, or executes some tasks in parallel; 5. During the continued reasoning process, if TA (accuracy) drops, the system calls the backup reasoning module or the score enhancement module; 6. Dynamically adjust until the TP_Score reaches the set threshold again to ensure the quality of the output result.

[0096] Supplementary description of the system's intelligent evolution and adaptive enhancement capabilities The AIOS task management operating system proposed by the present invention not only constructs a controllable task guidance system, but also reflects the development characteristics of "self-evolution" and "intelligent enhancement" in the system capability design. The core of this ability is based on the dynamic scheduling mechanism of the TP thinking model, enabling the system to real-time identify reasoning bottlenecks, cognitive biases, and path deviations during task execution, and automatically call the score enhancement module TI for supplementary reasoning.

[0097] In addition, AIOS provides an operating paradigm of "dynamically schedulable task thinking resource guidance system". By collecting behavior feedback through the UBB module and driving path reconstruction with the TP model score, it realizes the evolution ability from a static execution system to an intelligent feedback system, and has the system architecture characteristics of sustainable optimization, autonomous adjustment, and intelligent evolution. This concept is reflected in that the system not only supports multi-module collaborative control, but also has a self-optimization mechanism based on cognitive quality.

[0098] Description of the intelligent evolution control module To enhance the system's main control and scheduling ability for the reasoning path structure during task execution, the present invention adds an intelligent evolution control module.

[0099] The intelligent evolution control module is located between the task input parsing and module scheduling, and undertakes the functions of forming and optimizing the structure control language (TML) and path priority management. This module receives the scoring feedback from the TP model and the behavior preference data from the UBB service behavior database, and dynamically constructs multiple execution paths and schedules their priorities according to the scoring weights, task feedback data, and policy rules. Among them, the policy rules include the system-built path preference rules (such as preferentially selecting the path with the highest score based on the TP score), user preference guidance strategies (such as preferentially executing the fast branch path in combination with UBB data), and task history feedback adjustment strategies (such as rearranging the execution priorities according to the CSI score results). This policy set can be set and dynamically updated by developers, constituting the core basis for the system path scheduling.

[0100] The intelligent evolution control module has the following key functions: 1. Path combination construction ability: Automatically construct various combinations such as main paths, alternative paths, and innovative paths; 2. Execution order optimization mechanism: Dynamically adjust the path execution order in combination with the TP / CSI scoring feedback; 3. Scheduling center control logic: Coordinate module task invocation, resource allocation, and path reflux strategies.

[0101] As the core scheduling module for the path strategy evolution of the AIOS system, the intelligent evolution control module cooperates with the TP and TI modules to construct a complete intelligent inference chain control mechanism, significantly improving the system's adaptability and path evolution ability.

[0102] Description of the user evolution data management module As an intelligent evolution scheduling support component embedded in the AIOS system, the user evolution data management module is specifically used to long-term store, schedule, and manage the behavior data and path strategy data during the system task evolution, covering user behavior evolution information (output by the local guidance drive module) and path optimization scheduling data (output by the intelligent evolution control module). Its functions are as follows: 1. The user evolution data management module creates a unified data storage platform through a built-in lightweight local database (such as SQLite), manages multiple sub-table structures, including the user behavior log table (user_behavior_log), path combination record table (path_evolution_log), system space status table (meta_db_status), etc.

[0103] 2. The module has the ability to monitor the evolution data storage. When the database space reaches the set threshold, it automatically prompts the user for cleaning operations, and regularly executes the redundant strategy and invalid record cleaning mechanism to ensure the persistence and efficiency of the system operation.

[0104] 3. The module implements an access control mechanism for all evolutionary data. Users can view their historical evolution records, but have no right to directly edit or delete the original records, ensuring the objectivity and traceability of the intelligent evolution results.

[0105] 4. This module coordinates the data writing and reading between the local boot drive module and the intelligent evolution control module. As the backend support module for the execution of intelligent scheduling strategies, it ensures the continuous optimization of the system's long-term intelligent behavior, the stable migration of user preferences, and the gradual evolution and accumulation of task policy paths.

[0106] The user evolution data management module effectively makes up for the lack of a persistent learning and memory mechanism in existing AI task scheduling systems, enables the AIOS system to have the ability of long-term intelligent accumulation, provides a personalized evolution support path for large models, and significantly improves the intelligent task execution efficiency and adaptability of the system, with significant innovation and practical value.

[0107] Description of the Task Input Completeness and Output Expectation Matching Control Module To ensure that the AIOS system can obtain complete and accurate task input information from users during task execution and ensure that the AI output results highly match user expectations, the system introduces the task input completeness and output expectation matching control module.

[0108] This module guides users to fill in core fields such as task name, task description, target requirements, and background information at the task initiation stage through natural language interaction and structured input templates. At the same time, the system embeds an intelligent prompt mechanism to automatically judge whether input supplementation or target update is required at key nodes during task execution and generate appropriate prompts to assist users in improving task expression and constructing a clear structure of the "user intention diagram".

[0109] The task input completeness and output expectation matching control module is linked with the TP thinking ability evaluation mechanism, the task execution feedback module, and the system intelligent evolution control module to form a closed-loop input-output feedback chain. Through the real-time evaluation and source correction of output deviation, this module establishes a dynamic optimization mechanism centered on tasks and anchored by user expectations within AIOS. Its design is one of the key technical paths to achieve a "high matching comprehensive solution" and is also a basic guarantee module for the autonomous optimization of AIOS and the alignment with user cognition.

[0110] The task input completeness and output expectation matching control module and Figure 2 The TE-Control module in it jointly construct a task scheduling closed-loop, which is a key component for realizing task structure evolution and model path regeneration, and also constitutes one of the essential differences between AIOS and other static AI control systems.

[0111] The present invention constructs a dual thinking ability evaluation model system composed of TP and TI. The TP model conducts weighted scoring through three indicators: task thinking depth TD, thinking speed TS, and thinking accuracy TA, to form a basic score TP. When the system identifies a structural bottleneck or insufficient scoring, the TI model is dynamically activated to generate an increment ΔTI for enhancing the score, which supplements the original score. Finally, an enhanced score TP′ = TP + ΔTI is formed. The TP′ score result is input into the TE-Control module to drive path judgment and task scheduling. TP provides a stable main score, and TI provides structural enhancement and correction, constituting an intelligent coordination mechanism for scoring and execution. The cognition of the CSI module is not general cognition, but is embodied as: whether the AIOS can accurately and efficiently mobilize the thinking ability resources (the three elements of TA / TD / TS) of the AI model itself based on the nature of the user's task, and finally generate a comprehensive solution that meets the user's expectations. CSI is precisely the feedback bridge for this process. (See Figure 2 and Figure 6 ).

[0112] Example 5: Thinking Ability Enhancement and Dynamic Scoring System: This example details the design and implementation method of the thinking ability enhancement and dynamic scoring system in the AIOS system.

[0113] To further optimize the reasoning quality, enhance the cognitive depth of the system, and improve the creative reasoning ability, the present invention designs a thinking ability enhancement mechanism based on the TP basic thinking ability model, constructs a dynamic scoring system throughout the entire reasoning process, and realizes real-time adaptive control during the reasoning process. See Figure 6 , which shows the coordination relationship between the basic scoring model TP and the thinking enhancement module TI, and how to drive the task path judgment and the execution process of the language generation module through the conditional scoring enhancement path (TP′). The TP model conducts comprehensive weighted scoring based on three dimensions: task depth (TD), execution speed (TS), and thinking precision (TA), as the main basis for task execution scoring. If the TP score is insufficient, the system will dynamically activate the TI module to generate a structural enhancement score ΔTI, which is fused with TP to form an enhanced score TP′, as the conditional input for the path judgment logic. It should be particularly noted that the TI module and the ΔTI path are not called for every task scoring. The generated TP′ score belongs to the "conditional enhancement path" and is only triggered when the TP score is insufficient to achieve an intelligent scoring reinforcement mechanism. The scoring results of TP or TP′ are used by the TE-Control module for path decision-making to guide the Prompt generation module to generate a structural or innovative output task path.

[0114] (1) The specific design of the thinking ability enhancement mechanism is as follows: When the system detects insufficient reasoning depth, the reasoning path getting stuck in a bottleneck, or a decline in accuracy, it automatically invokes the scoring enhancement module (TI module) to inject control language containing alternative path suggestions, non-linear thinking parameters, or multi-round mutated Prompts into the AI model, thereby guiding the AI model to generate more creative new paths or complete the reasoning chain.

[0115] The cognitive increment generated by the scoring enhancement module is received and evaluated by the TP scoring system to form an enhanced score TP′, achieving an intelligent improvement in the system's output ability.

[0116] (2)TP′ (Enhanced Thinking Ability Score) Formula The system thinking ability enhancement mechanism introduces a cognitive increment ΔTI to correct and enhance the original TP score, forming the TP′ score: TP′ = TP + ΔTI Among them, TP = λ1·TD + λ2·TS + λ3·TA is the original scoring structure, and ΔTI is dynamically calculated by the TI module during path generation, reflecting the scoring bonus of the newly added cognitive value.

[0117] In the extended structure, it can also be expressed as: TP′ = λ1·TD + λ2·TS + λ3·TA + λ4·ΔTI, where λ1+λ2+λ3+λ4=1 When the TI module successfully introduces an increment in thinking ability reasoning, the system adjusts the TD and TA scores according to the influence degree of the increment to form a new scoring standard for TP′; the more effective the TI is, the more obvious the increase in the TP′ score, thus encouraging the system to boldly invoke thinking ability resources when necessary.

[0118] (3)The dynamic scoring system runs throughout the whole process The dynamic scoring system runs continuously during reasoning execution and mainly includes the following functions: Real-time scoring monitoring After each reasoning stage is completed (such as when a reasoning path or a task module execution is completed), the system immediately updates the TD, TS, TA, and TI indicators and calculates the new TP′ score.

[0119] Scoring trend analysis It not only monitors the single-point score but also analyzes the scoring change trend to judge whether the reasoning process is continuously optimized, stagnant, or deteriorating.

[0120] Adaptive reasoning adjustment According to the TP′ score and the result of trend analysis, dynamically adjust the reasoning depth, path complexity, frequency of alternative path invocation, and activation frequency of the scoring enhancement module.

[0121] Output accuracy control The TP' score also serves as the basic basis for the subsequent T-Level output control module (see Embodiment 6), directly affecting the depth and integrity of the output content.

[0122] (4) Example process of enhanced thinking ability 1. Calculate TD, TS, and TA in real time during the reasoning process; 2. When it is detected that the TD score of the reasoning depth decreases, activate the TI module; 3. The score enhancement module generates a new auxiliary reasoning path to supplement the reasoning depth; 4. Incorporate the TI increment into the TP' score and update the reasoning state; 5. Dynamically adjust the reasoning strategy to ensure that the final output meets the requirements of cognitive depth and accuracy.

[0123] Embodiment 6: Application of the T-Level output control module: This embodiment details the design and application method of the cognitive output level mechanism of the T-Level output control module in the AIOS system. For details, see Figure 7 , which shows the complete operation logic and control process of the AI intelligent task management operating system embedded with the thinking ability evaluation model. The system starts with the task goal input by the user, goes through the structured information extraction and NLP semantic parsing module, and enters the initial stage of the execution path. Inside the system, the TP thinking ability evaluation model comprehensively evaluates the complexity, depth, speed, and accuracy of the task, and decides whether to call the TI thinking ability enhancement module for structure supplementation according to the "satisfaction judgment path". The core of the system is driven by the TE-Control module, supporting multi-round dynamic task scheduling and optimal path selection. In the execution path, the system can automatically switch between the ordinary task processing and innovative task processing mechanisms to ensure that the generated Prompt structured prompt meets the user's expectations. Figure 7 It reflects the whole-process closed-loop of the AI task system from language parsing to thinking ability evaluation, then to decision control and prompt generation, emphasizing the adaptive ability and innovation ability of the system.

[0124] In order to adapt to different task complexities, user requirement levels, and cognitive depth requirements of reasoning results, the present invention proposes a T-Level output control module driven by the TP' score to achieve dynamic hierarchical control of reasoning output and adjustment of cognitive depth. For details, see Figure 8, which shows the overall process logic of the structured task output and the user configuration mechanism. The system first receives the user task input content, including the task objective and specific requirements. Subsequently, through the AIOS-3W user retrieval control mechanism module and the T-Level output control module, it conducts phased analysis of the task objective and matches the output level. On this basis, the system constructs a PSCP output model, and according to different task scenarios and user target types (such as product design, strategic planning, research projects, etc.), generates multi-dimensional structured content, including key items such as product design, function introduction, design scheme, cost calculation, etc., and presents and adapts them according to different levels (L1~L4). Finally, the system outputs a comprehensive solution that meets the user's expectations, achieving an accurate correspondence and quality guarantee between the task content and the target requirements.

[0125] (1)Hierarchical System Design of the T-Level Output Control Module The T-Level output control module is a module in the system for hierarchical scheduling of structured output. It controls the output path and filters the results based on cognitive priorities, serving as the execution carrier of the cognitive output level mechanism; the T-Level output control module divides the task output into different cognitive depth levels, and each level represents an improvement in aspects such as logical integrity, reasoning details, and analysis angles of the output content. The standard division is as follows: T1 level: Quick response output Based on preliminary reasoning, provide a quick but coarse-grained answer. The output speed is fast and is suitable for simple tasks or real-time feedback requirements.

[0126] T2 level: Standard reasoning output On the basis of T1, add logical details and conditional analysis. The output content is more complete and is suitable for tasks of regular complexity.

[0127] T3 level: Deep reasoning output Combine multi-angle analysis and auxiliary reasoning paths. Suitable for high-complexity tasks or scenarios requiring comprehensive analysis.

[0128] T4 level: Multi-dimensional comprehensive reasoning output Introduce the reasoning results of the scoring enhancement module (TI). The output covers predictive analysis, hypothetical deduction, and innovative deconstruction, and is suitable for decision support and complex system modeling.

[0129] T5 level: Output with extremely high cognitive depth Combine historical task evolution data (UBB module), multi-round feedback adjustment, and innovative reasoning to generate the optimal comprehensive solution in all dimensions. Suitable for fields with extremely high cognitive requirements such as high-end think tank analysis and complex strategic planning.

[0130] (2)T-Level Judgment Basis The T-Level hierarchy is dynamically determined based on the following three major factors: The higher the TP' scoring level TP', the more capable the system is of generating higher-level outputs; Task complexity label The task marks its complexity during the parsing stage, and high-complexity tasks default to pointing to high T-Level outputs; User output requirement setting The user can specify the desired T-Level during the task input stage, or the system can automatically infer the user's preferences through historical interaction behaviors (UBB database).

[0131] (3) Output generation and control process 1. The system monitors the change of the TP' scoring during the reasoning process in real time; 2. Considering the task complexity and user requirements setting comprehensively, determine the current target T-Level; 3. According to the T-Level standard, control the logical depth, detail richness, and number of reasoning branches of the output content; 4. If the TP' scoring drops during the reasoning process, automatically downgrade the T-Level to ensure the timeliness of the output; 5. If the reasoning quality improves or the thinking increment becomes richer, automatically upgrade the T-Level to provide higher-quality output.

[0132] (4) T-Level application example Task example: Formulate a digital transformation plan for medium-sized enterprises T1 output: Give a general process list of digital transformation; T2 output: Combine the current situation of the enterprise and list the key steps and precautions; T3 output: Analyze the current situation of each department of the enterprise and formulate detailed phased transformation goals; T4 output: Introduce cases of innovative transformation models and predict potential risks and opportunities; T5 output: Based on the enterprise's historical data, industry trends, and innovative thinking results, formulate a dynamic and adjustable strategic blueprint.

[0133] Example 7: Evolution of the user business behavior database (UBB) and local cognitive adaptation: This example details the design of the user business behavior database (UBB) module in the AIOS system and the specific implementation method of the local cognitive adaptation evolution mechanism. For details, see Figures 9 - 10 , Figure 9 For the local AI cognitive execution joint control system logic diagram, the system parses, schedules, and intelligently responds to the task content through the collaborative execution system of the AI large model deployed locally and the AIOS task operating system. Figure 9The core mechanism revolves around the TP thinking ability evaluation model. The system evaluates the depth, speed, breadth, and accuracy of tasks through the TP model, and on this basis, the CSI cognitive scoring and feedback module further identifies the user's intention and output accuracy. Among them, the CSI cognitive scoring and feedback module generates a cognitive score through the combined weighting of three key dimensions (PMR, MIS, OHR). Inside the system, a linkage mechanism is formed by the Prompt generation module, Transformers execution module, TE-Control module, and T-Level output control module to ensure the precise matching and quality control of the output content. At the same time, the system can integrate the output behavior data through the output integration module and write it into the UBB user business behavior database to form a feedback loop for subsequent optimization iteration and local training support. The overall process emphasizes the deep integration of the local deployment model and the AIOS system to ensure data security, real-time response, and model controllability, and finally generates a comprehensive solution that meets the user's expectations. Figure 10 It is a local adaptive training logic diagram driven by user business behavior. It shows the AI local adaptive training logic mechanism driven by user business behavior. Based on the user task input, the system performs semantic parsing and TP thinking ability scoring through the AIOS task management system, and conducts a feedback evaluation on the task execution effect through the CSI cognitive scoring and feedback module. During the execution process, the system retrieves the local AI model in real time, dynamically adjusts the task execution instructions through the Prompt generation module and the Transformers execution module, and continuously writes the task behavior data and system response data into the UBB user business behavior database. The local system continuously collects user business behavior data, establishes a cognitive model of the user business behavior characteristics through the cognitive behavior model generation module, and realizes personalized adaptation of task scheduling, output preference, and inference path based on this model, thereby improving the accuracy of system response and user satisfaction and forming a cognitive behavior data closed-loop. Figure 10 The structure in the lower right part shows that the system performs operations such as "task feature modeling", "cognitive evaluation model generation", and "behavior data comparative analysis" based on the local UBB data, and injects the optimized operating parameters back into the AIOS system to form a local self-learning and self-optimizing intelligent training channel. This mechanism ensures that the AIOS system has the ability of adaptive adjustment, can continuously evolve the model performance according to the actual business behavior of users, and improve the response accuracy and task completion ability of AI in specific enterprise or user scenarios.

[0134] To improve the system's cognitive adaptation ability to individual user needs and realize the dynamic evolution of task execution strategies, the present invention introduces the UBB (User Behavior Base) user business behavior database module, which drives the system to perform cognitive adaptive optimization in the local environment by continuously recording, analyzing, and applying user interaction behavior data.

[0135] (1)Design of User Behavior Database (UBB Module) The UBB module is specifically used to collect, store, and manage the following types of user behavior data: Task input records The task theme, description, objectives, and parameter information submitted by users.

[0136] Task execution feedback The evaluation, correction suggestions, and task reconstruction information output by users to the system.

[0137] Operation behavior data Including the user's interface operation trajectory, option selection, path preference, etc. during task execution.

[0138] Cognitive preference information Through multiple rounds of interaction processes, the system automatically infers the user's preference settings for reasoning depth, response speed, and output accuracy.

[0139] (2)Data Collection and Privacy Protection When collecting data, the UBB module follows the following design principles: Principle of minimum necessity: Only collect the data necessary to complete cognitive adaptation, without accumulating redundant data.

[0140] Local storage priority: Store behavior data in the user's local device or dedicated server first to ensure privacy security.

[0141] User controllability: Users can view, export, or clear their own behavior data records at any time.

[0142] (3)Local Cognitive Adaptation Evolution Mechanism Based on the behavior data accumulated by the UBB module, the AIOS system can execute the following adaptive optimization processes: Cognitive preference modeling By analyzing historical task execution behaviors, establish a user's cognitive preference model (such as preferring concise answers or detailed deductions, preferring speed priority or accuracy priority).

[0143] Adaptive adjustment of reasoning parameters Before the execution of a new task, dynamically adjust the reasoning parameters according to the user's cognitive preferences, including: the default output level of T-Level; the complexity of the reasoning path; the invocation frequency of the scoring enhancement module (TI module); the dynamic feedback weight allocation.

[0144] Personalized evolution of execution strategies The AIOS system continuously fine-tunes the reasoning process and module invocation order according to the behavior evolution data, and gradually forms an execution style and cognitive logic that highly fits the specific user needs.

[0145] Cognitive evolution feedback closed-loop After each task execution is completed, the execution effect and user feedback are written into the UBB database again to form a self-evolving cognitive closed-loop, and the system's cognitive adaptation ability is continuously improved.

[0146] (4)Example application process 1. The user completes multiple task submissions and output feedback; 2. The system analyzes the behavior data and identifies that the user tends to output with high cognitive depth (prefers T4 and T5 levels); 3. During the execution of subsequent new tasks, the system automatically increases the reasoning depth and preferentially introduces an enhanced thinking module; 4. After the task output, continue to collect user feedback to refine the cognitive preference model; 5. As the number of uses increases, the system is highly synchronized with the user's cognitive style, improving the output compliance and user satisfaction.

[0147] Example 8: AIOS and AI model collaboration system: Example application This example details the implementation method and example application of the collaboration mechanism (i.e., the cognitive collaboration system) between the AIOS system and the AI model.

[0148] To overcome the problems of insufficient cognitive reasoning depth and limited autonomous adjustment ability of traditional AI models in complex task execution, the present invention forms a close collaboration between the AIOS system's structured input, dynamic guidance of thinking ability, cognitive evolution mechanism and the underlying AI model to achieve a non-linear leap in overall cognitive intelligence.

[0149] (1)Overall design of the collaboration system When the AIOS system collaborates with the AI model, the division of labor is as follows: AIOS system: Responsible for task parsing, reasoning structure construction, dynamic guidance of thinking ability and feedback optimization; manage the dynamic adjustment during the reasoning process, trigger enhanced thinking and cognitive evolution.

[0150] AI model (underlying model, such as DeepSeek, GPT, etc.): As a cognitive execution resource, it completes specific reasoning, inference and content generation according to the control instructions (TML instructions) generated by the AIOS system; responds to the module calls, reasoning adjustments and enhanced thinking trigger requests of the AIOS system. The two collaborate to form a complete closed-loop, which is no longer a one-way instruction input, but a structured cognitive process control and dynamic reasoning interaction.

[0151] (2)Collaboration operation process 1. Task input stage The user inputs task requirements through natural language + logic diagram language, and the AIOS system parses and generates a standardized inference structure.

[0152] 2. Inference Preparation Stage Based on the task structure, the AIOS system formulates an inference path, a control instruction set (TML), and sets dynamic feedback monitoring points.

[0153] 3. Inference Execution Stage The AI model gradually completes the inference execution according to the TML instructions, and the AIOS system monitors the inference depth, speed, and accuracy in real time.

[0154] 4. Dynamic Adjustment Stage If inference deviations or bottlenecks are detected, the AIOS system dynamically adjusts the inference path, switches modules, or invokes enhanced thinking functions to guide the AI model to continue reasoning.

[0155] 5. Cognitive Output Stage After the inference is completed, the AIOS system determines the output T-Level hierarchy according to the TP′ score, and controls the output precision and depth.

[0156] 6. Feedback Learning Stage The user feedback information and task execution data are written into the UBB user business behavior database for optimizing the inference strategy before the next inference.

[0157] Through the above process, the AIOS system forms cognitive structure control and inference optimization guidance for the AI model, enabling the overall cognitive intelligence level to exceed that of a single AI model; (3)Example Application Scenarios Example 1: Financial Risk Control Strategy Formulation Traditional Mode: Directly ask the AI model for risk control strategies to obtain single suggestions or fixed processes.

[0158] AIOS Collaboration Mode: Analyze the specific financial business background and risk indicator requirements input by the user; dynamically infer and generate multiple alternative paths for risk control strategies; adjust according to real-time scoring and screen the optimal risk control combination; introduce a scoring enhancement module to generate a predictive risk evolution model; output a multi-dimensional and dynamically adjustable risk management plan, with T4 / T5 level output.

[0159] Example 2: Intelligent Medical Assistant Decision-making Traditional Mode: The AI model directly gives standardized treatment suggestions.

[0160] AIOS Collaboration Mode: Analyze the patient's medical records, treatment history, and individual difference characteristics; construct a multi-path diagnosis and treatment reasoning structure; dynamically adjust the treatment plan combination, and optimize the treatment suggestions according to real-time data feedback; introduce innovative treatment ideas and explore the possibility of new therapies; output a comparative analysis of multiple plans and dynamically adjusted recommendations, providing high-cognitive-depth support for doctors' decision-making.

[0161] (4)Summary of the advantages of the collaboration system Improve the quality of task analysis and inference structure generation; enhance the dynamic adaptability and innovation of the inference process; significantly improve the cognitive depth, logical integrity, and user fit of the final output; form a truly human-machine collaboration and cognitive evolution intelligent system.

[0162] Supplementary note: Deployment method and interaction interface adaptation of the AIOS system The deployment strategy of the AIOS system fully embodies the design concept of "minimal intrusion and maximum activation". In the entire system architecture, AIOS does not replace the original functional interfaces or interaction channels of the AI model. The present invention designs a model adaptation guidance module to guide the system to adapt and connect with different types of AI models after the task structure is generated. Through the structure conversion and input reconstruction mechanism, the universality and low intrusion of model access are realized, providing a unified access channel for supporting multi-model deployment (including open-source models and private models), and it is a bridging module for converting the task execution logic into an AI model; the present invention dynamically activates the thinking power resources inside the AI model and schedules its native inference ability in the way of an embedded guidance mechanism, so as to output a comprehensive solution that better meets the user's expectations.

[0163] Therefore, the AIOS system can be directly and seamlessly integrated with open-source AI large models (such as DeepSeek, ChatGLM, Mistral, etc.) during deployment, without developing a new graphical user interface (UI). The system is recommended to use the native interaction interface provided by the AI model (such as Web UI or CLI) to complete the interaction and execution process of user tasks at the prompt structure and thinking path control levels.

[0164] This method simplifies the system deployment path of AIOS and improves the system operation stability on the one hand; on the other hand, it also fully reflects the independence of the AIOS system: its core protection is an intelligent task guidance system based on the "Prompt+Steering+TP evaluation mechanism". The present invention mainly focuses on the method path and scheduling mechanism, without involving changes to the UI or the model structure itself.

[0165] The AI task operating system proposed by the present invention provides controllable, secure, and intelligent collaborative underlying support for future intelligent work platforms and new quality productivity infrastructure through the integrated application of a task control system, a guided open-source AI model platform, and an enhanced computing environment, and has broad potential for personalized applications and social promotion value.

[0166] In summary, the AIOS system can enhance the functions of intelligent task management and inference guidance without building an independent UI portal, which conforms to the core idea of "guided intelligent enhancement" of this system.

[0167] The system of the present invention has closed the scoring and scheduling loop and has the ability of self-evolution. The specific index descriptions are shown in Table 1: Table 1

[0168] The above embodiments are used to explain the present invention, rather than limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims of the present invention fall within the protection scope of the present invention.

Claims

1. An AI task operating system with a dual model for evaluating thinking ability, characterized in that, The system includes: A task input module, which is used to receive the natural language description and / or the logical graph language structure input by the user, and generate standardized task input information; An NLP semantic parsing module, which is used to perform semantic parsing on the natural language input, extract task objectives, logical relationships, and execution conditions, and fuse and parse them with the logical graph structure to generate a unified standardized task input structure; A task scheduling engine module, which is used to generate control language instructions according to the parsed task structure and construct an inference path; A thinking power scheduling and feedback optimization module, which is used to comprehensively evaluate the thinking depth TD, thinking speed TS, and thinking accuracy TA based on the thinking power evaluation model TP, and dynamically adjust the inference path and module calls according to the scoring results; A structured output processing module, which is used to generate output content with different precisions and cognitive depths according to the cognitive output results after the inference is completed; A user behavior database module, which is used to record the operation behaviors, feedback information, and cognitive preference data of the user during the task execution process; A cognitive collaboration module, which is used to collaborate with the underlying AI model to execute task inference and cognitive output through structured input, inference control, and dynamic feedback optimization.

2. The AI task operating system according to claim 1, wherein: The task input module further includes: A natural language input module, which is used to receive the task description information expressed by the user in natural language, and extract the task objective, logical relationship, and execution condition through the NLP semantic parsing module; A logical graph language input module, which is used to receive task nodes, execution paths, control logic, and parameter conditions expressed in a standardized logical graph form.

3. The AI task operating system according to claim 1, wherein: The task scheduling engine module further includes: A control language generator TML, which is used to generate internal control language instructions according to the standardized task input structure, and the control language instructions define task node calls, execution order, condition judgment, exception handling, and alternate path setting; An inference path construction module, which is used to construct an inference path according to the control language instructions, and the inference path includes a main inference path, an alternate inference path, and corresponding path weights are assigned.

4. The AI task operating system according to claim 1, wherein: The thinking power evaluation model TP in the thinking power scheduling and feedback optimization module includes: A thinking depth TD score, which is used to evaluate the hierarchical depth of task node calls and the tightness of the inference logic during the inference process; A thinking speed TS score, which is used to evaluate the time consumption of the inference path execution and the inference response efficiency; A thinking accuracy TA score, which is used to evaluate the consistency between the inference output result and the task objective or user expectation; A dynamic weight adjustment mechanism, which is used to dynamically adjust the weight coefficients of the TD, TS, and TA indicators according to the task type, execution stage, or user preference, so that the total thinking power score TP_Score optimally matches the task execution requirements; Among them, the total thinking power score TP_Score is calculated according to the following formula: TP_Score = λ1×TD + λ2×TS + λ3×TA In the formula, λ1 + λ2 + λ3 = 1, which are the weight coefficients corresponding to the dynamically adjusted TD, TS, and TA; A dynamic adjustment mechanism is used to dynamically adjust the flow of the inference path during task execution according to inference feedback results, changes in thinking ability scores, or abnormal trigger conditions, including switching to alternative paths, reallocating inference resources, or adjusting the execution order.

5. The AI task operating system according to claim 1, wherein: The thinking ability scheduling and feedback optimization module further includes: A thinking ability enhancement mechanism. The thinking ability enhancement mechanism introduces a thinking increment ΔTI based on the scoring enhancement module TI and incorporates the thinking increment into the total thinking ability score to form an enhanced thinking ability score TP′, where: When insufficient inference depth, decreased inference accuracy, or the inference path getting stuck is detected during the inference process, the TI module is automatically invoked; The TI module generates new auxiliary inference paths, alternative inference schemes, or inference branches, sets the corresponding thinking increment ΔTI as supplementary cognitive value for inclusion in the adjustment of the thinking ability score; The calculation formula for the enhanced thinking ability score TP′ is: TP′ = TP + ΔTI where TP represents the original total thinking ability score TP_Score, and ΔTI, as an enhanced scoring factor, participates in the dynamic adjustment of the weight coefficients of each index of TD, TS, and TA and path optimization in real time during task structure recognition or path inference in combination with the system task complexity and historical task scoring trends, constituting a dynamic enhancement mechanism for the system thinking ability score TP′.

6. The AI task operating system according to claim 1, characterized in that: The structured output processing module further includes a T-Level output control module. The T-Level output control module dynamically determines the cognitive depth and precision level of the output content according to the thinking ability score TP′ results, task complexity, and user requirements. The T-Level output control module includes: Level T1: Quick response output based on preliminary inference results, with concise output content and priority given to response speed; Level T2: Adds logical details and conditional analysis on the basis of T1 to provide standard inference output; Level T3: Generates in-depth inference output by combining multi-angle inference and auxiliary path analysis; Level T4: Introduces new inference content generated by the scoring enhancement module TI to form a multi-dimensional comprehensive inference output; Level T5: Outputs a comprehensive analysis scheme combining in-depth inference and multi-dimensions based on historical data, feedback evolution, and inference results; where the T-Level output control module dynamically upgrades or downgrades the output level according to the TP′ score during task execution to adaptively optimize the task output quality and response speed.

7. The AI task operating system according to claim 1, characterized in that: The user behavior database module is used to record the input content, execution feedback, operation behavior trajectory, and cognitive preference information of the user during task execution, extract the personalized preference characteristics of the user for inference depth, inference speed, and output granularity, and use these preference characteristics for task output control and execution path optimization.

8. The AI task operating system according to claim 1, characterized in that: The cognitive collaboration module is used to collaborate with the AI model to perform task inference and cognitive output, specifically including: A structured inference control mechanism for transmitting standardized task inputs, inference paths, and dynamic feedback instructions to the AI model through the control language TML to guide the AI model to perform inference tasks; A dynamic thinking ability guidance mechanism, which is used to adjust the inference path, switch to an alternative path, activate the scoring enhancement module according to the real-time TP' score or TP_Score and the inference feedback, and transmit the adjustment instructions to the AI model; An innovative inference collaboration mechanism, which is used to call the innovative inference ability of the AI model during the inference process to generate new inference branches or auxiliary solutions; A cognitive feedback optimization mechanism, which is used to dynamically optimize the inference strategy according to the inference execution results and user feedback, and continuously improve the overall cognitive inference effect.

9. The AI task operating system according to claim 1, wherein: The control language generator TML includes a logic graph language control module, specifically as follows: (1) The graph language task node structure TML-Node, which represents the structural actions or judgment units of the task; (2) The graph language path connection structure TML-Path, which represents the logical jump and execution order between task steps; (3) The graph language control symbol structure TML-Control, which is used to express conditional judgment, exception handling, and path switching; (4) The graph language parameter label structure TML-Tag, which binds task parameters, user expectations, and output targets.

10. The AI task operating system according to claim 1, wherein: The AI task operating system, together with the personal PC and the open-source AI model platform, constitutes a three-in-one intelligent execution architecture; The AI task operating system is responsible for task input parsing, thinking ability scheduling, control path generation, and feedback closed-loop control; The personal PC has local computing power resources, a structured task execution memory space, and a task management and scheduling unit; The open-source AI model platform conducts two-way communication and task takeover with the AI task operating system through the interface protocol.

Images