AIP Institute buckets work towards developing a neural network content unit AIP for use in cooperation with the AI on the basis of large language models (LLM) as GPT-3 (OpenAI), BERT (Google AI), YaLM (Yandex).
The main features of such neural network information blocks of AIP are the possibility of storing controlled information structures in the AIP block and using neural network technologies to obtain multiple representations of information in a limited context of knowledge verified by specialists. Unlike general-purpose multitasking AI, which are rather aimed at the task of responding to a user’s request in “any case” and having problems with preserving unique weakly representative knowledge, our development is aimed at preserving and controlling information consisting in responses to the user in a specific subject area, such as education, corporate systems or for human-machine interaction systems. In such information transmission systems, an important component is compliance with standards and verification by experts of the output that the user receives when interacting with AI, which is currently considered one of the problems of open AI as ChatGPT. The developed technology of a domain-oriented small neural network language model also allows generating various representations of information, taking into account the parameters of the user profile and the capabilities of the end device for evaluating and transmitting information.
Goals of using AI technologies
- Full or partial automation of the information structuring process based on unstructured data sources for use in information transmission systems based on AIP methodology.
- Optimization of the process of generating various representations of information (knowledge) in a limited context of the subject area.
- Creation of a recommendation system based on the analysis of graph structures of the user profile and the general Ontology of cognitive data (Knowledge Base)
- Development of methods for assessing the current state of the user to highlight the personalization parameters of the information transfer process, depending on the available knowledge, physical capabilities and capabilities of the end devices of information transmission.
Tasks solved using AI technologies and ADANEC tools
Intelligent analysis of unstructured information (documentation, educational manuals, etc. sources) by subject area for:
- Context generation for blocks of scenarios for transmitting information to AIP within a given subject area using a selection of documents;
- Creating an ontology of the subject area based on the allocation of named entities, facts, key concepts in the form of keywords and phrases;
- Establishing semantic links between entities in the ontology of cognitive data based on context analysis from a set of documents on a given subject area and assessing the proximity between key concepts (keywords/phrases), named entities, fact-checking;
- Converting data formats to from Text to Image (Text-to-Image), Text to Video (Text-to-Video), Video to Text (Video-to-Text), Image description (Image-to-Text).
Creation of various representations of the context in information transmission systems, both for users and for intersystem interaction (for example, sending commands to a robot):
- Generation of a sequence of blocks of information in the form of scripts for transmitting information to a user or other information system;
- Generating test tasks based on the context of the subject area;
- Generation of instructions in the form of a sequence of actions required to be performed by the user.
Creation of visual forms of information representation in user interfaces of various types:
- Generation of visual representation of information blocks of scripts (content) in the form of executable code with specified parameters of the interface for displaying/transmitting information;
- Generation of forms for testing and evaluating knowledge in the limited context of an information block of a scenario or a set of blocks;
- Generation of illustrations, graphs and video content based on the context of information blocks of scenarios.