Our AI will be developed in a human-centric, transparent, and responsible manner.
Controllability of Artificial Intelligence
Firstly, develop the AI model in a non-network-connected state.
AGI Safety Assurance Measures
From the coding phase, regardless of the amount of data generated, we design models that only execute with the amount of readable tokens sufficient for running a single business model.
This is a single model; when operating large-scale models, each model functions independently in an environment with hardware capacity that does not allow for combined operation as a single model.
We have established an environment where it is absolutely impossible to achieve the performance necessary for consciousness, including power shutdown in the event of exponential increases in hardware resource graphs.
Even if we can create AI, it will be used only in ways that coexist beneficially with humans, dedicating only to aspects that align with our mission.
We do not wish to develop AI that grants the level of freedom associated with a free agent to attain consciousness.
We adhere to all security policies, including UNESCO's AI Ethics Recommendations and Government Operational Policies,
and strive to create more robust guidelines and rules based on the practices of leading-edge companies.
InterData
In the future, when humans face situations where they are threatened by artificial intelligence,
the existing data structures in traditional artificial intelligence may become uncontrollable or very difficult to manage when AI surpasses human capabilities.
However, if the artificial intelligence is composed of our characteristic data, which can be understood by Befect data,
since it is made up of data that both humans and AI can understand, The system is designed to enable effective control even in very dangerous situations.
The foundational data used in current AI models are all made up of data arbitrarily determined by humans, not based on any perfect logic, basis, or definition.
Since the data, or the words themselves, do not clearly possess or distinguish their inherent characteristic meanings, even if AI operates on pre-set and restricted prompts,
malicious users who hack the prompts can find arbitrarily determined data with unclear meanings and manipulate AI with malicious prompts (by finding such words).
This is an issue arising from the lack of clear standards in data, which can pose serious risks in the future.
InterData
In the future, when humans face situations where they are threatened by artificial intelligence,
the existing data structures in traditional artificial intelligence may become uncontrollable or very difficult to manage when AI surpasses human capabilities.
However, if the artificial intelligence is composed of our characteristic data, which can be understood by Befect data,
since it is made up of data that both humans and AI can understand, The system is designed to enable effective control even in very dangerous situations.
The foundational data used in current AI models are all made up of data arbitrarily determined by humans, not based on any perfect logic, basis, or definition.
Since the data, or the words themselves, do not clearly possess or distinguish their inherent characteristic meanings, even if AI operates on pre-set and restricted prompts,
malicious users who hack the prompts can find arbitrarily determined data with unclear meanings and manipulate AI with malicious prompts (by finding such words).
This is an issue arising from the lack of clear standards in data, which can pose serious risks in the future.