Artificial Intelligence Virtual Machine

Artificial Intelligence Virtual Machine (AIVM)

AIVM artificial intelligence virtual machine is a problem solver (PS), including input and output sub-machine, knowledge sub-machine, multi-intelligent agent system (MAS) sub-machine, and decision-making system (DS) sub-machine. AIVM becomes the basis for application servers (AS) in various fields and provides APIs for them to call. The input and output sub-machine connects ChatGPT, Bard and BLOOM, and provides AIVM "services" for various application fields. Many companies or government agencies keep their data or knowledge confidential, and they do not want this data or knowledge to stay in OpenAI, Google, or public clouds. Therefore, AIVM uses the GPT connector of the input and output sub-machine to connect to these well-known large language models (LLM) on the one hand using their Programming Interfaces, and on the other hand to connect to the application server of the company or government agency.

The above process of processing private data or knowledge can be applied to both corporate groups and individual consumers. In the case of individual consumers, the applications processed by the application server are personal AI applications similar to mobile phones.

1 AIVM Problem Solver input and output slave (PSIO)

When the user asks, first enter the input encoder of the problem solver PS. After complex processing, the responses from each GPT are summarized, and finally the output encoder gives the answer to the user.

1.1 AIVM Complex Feedback System (CAS) and Input Encoder

Input to AIVM can take many forms. One of them is the traditional GUI. This part may be related to the application field, but AIVM provides more general GUI code, which customers can modify according to their own needs. Customers can also use Chat to gradually learn and obtain the SPEC of the problem as input through continuous chat interaction. For example, SPEC is learned through BLOOM's Chat function and BLOOM input encoder (IE). SPEC first entered CAS to expand BLOOM so that it has the genetic algorithm characteristics of Problem Solver. The operation of CAS must be connected to the AIVM knowledge submachine.

1.2 AIVM GPT Connector (GC)

GC first (1) connects to multiple LLMs, and after obtaining each answer (2) in GC, uses its own data or knowledge, as well as BLOOM's GPT algorithm, to summarize and correct the answers of LLMs, which must use MIT research. result. Then (3) receive the application server. (The order of 1-3 may need to be adjusted. For example, first use your own BLOOM and data to create the answer, and then summarize each LLMs to get the final answer)

1.3 AIVM output encoder Output Encoder (OE)

Finally, use BLOOM to create a response to the user.

2 AIVM knowledge submachine

This part is a combination of BLOOM GPT and application knowledge base/BFO, in order to prevent the hallucination problem of GPT.

3 AIVM Multi-Agent System (MAS) slave machine

Integrate Facebook's open-source Cicero, OpenAI's multi-agent-emergence-environment, and Google DeepMind's Protagonist Antagonist Induced Regret Environment Design (PAIRED).

4 AIVM Decision System (DS) slave

Open-source decision system using leverage model.

5 AIVM abstract submachine

The AIVM abstract submachine contains various system theories or principles.

6 AIVM basic slave machine

The basic slave machine provides the foundation of AIVM, which in turn is the foundation of the application server. Therefore, the basic AIVM submachine is the "basic foundation".

7 AIVM Application Server (AS)

For groups, this part can be developed in cooperation with several companies in various application fields. As for individual consumers, the applications processed by the application server are personal AI applications similar to mobile phones.