å
TL;DR
The AI renaissance has brought a swift and direct impact on the digital world. With Large Language Models taking over our timelines, podcasts and search engines the fact is, that everyone is incubating into the same limited set of centralised systems. This should be a worry for each citizen that utilises the uses of AI, built by centralised corporations.
The want for AI is only going to grow, the use cases will expand exponentially - along with its intelligence. To create true personalisation, companionship and most important, security.
PIA is the answer. A smaller language model is built to understand humans and to understand delegation and separation from the larger models. PIA is trained with no one at its centre but an objective. Deployed as a new instance for every user, PIA is able to self-teach, build and iterate according to the user's needs.
Terms
- LLM - Large Language Model
- LM - Language Model
- Dataset - A batch or information used to train an AI model.
Introduction
AI and AI-driven products are here to stay. The power, accessibility and efficiency that they bring will increase the general humans (who embrace them) efficiency and productivity not only in their work environment but also in their personal life. Large language models will serve as the baseline of all major general AI systems, allowing super-fast access to huge amounts of data. With more specific models being trained to provide highly specialised services.
The biggest removal from human interaction that these models will bring is a level of personal connection. The ability for a model to learn and grow with its uses, bringing the ability to truly connect, advise and enhance. With personal connection comes privacy and security. The first mainstream models to provide a connection with its human user will be the for-profit, multi-national corporations. This in itself brings concern for data privacy, as it will allow them to have the most detailed form of personal data collection we have seen in the digital world to date.
PIA give a solution by providing a secure and personal AI model that is trained to service only a single user or family. With security at its heart, PAI will be a small LM that is trained on how to service, communicate and connect. It will not be trained on a personality, allowing it to come to its users fresh and untainted. A small, self-teaching, self-reliant model that runs independently of any other outside systems. But at its core understands how to utilise third-party tools, like LLM’s, websites, Apps and APIs, allowing it to execute in any situation for its user. While keeping private data secure and abstract from any third-party Ai systems.
The Problem
Models will range in size and centralisation from open sources and private to corporate and closed. Private models, ones you run on your phone or laptop, will be the most secure but also the slowest learning and