What does ChatGPT mean to Personal AI?
In 2022, deep-tech startups faced a rough start. Our push for personal AIs for individuals, as well as ownership of data developed on web3 standards, was seen as TFT (too far too soon). We may have been early to write and execute on the thesis, but with the release of ChatGPT, the focus on AI emerged.
As the year came to an end, the question then changed from if personal AI is possible to how fast can I bootstrap my personal AI. We saw more consumers looking to train their own personal AIs as the awareness around ChatGPT penetrated into the consumer market.
Given the similar chat interface we created for Personal AI, we also encountered a common question in the minds of everyone at the same time: “What does ChatGPT mean to Personal AI?”
ChatGPT, as most of us now know, is a conversational layer that allows chat-like instructions and interactions with the GPT-3 system, which is similar to Personal AI chat interface or any other messaging-like interface. Personal AI models are the opposite of Open AI models in that a personal AI model is based on individual personal data (memory stack), rather than the internet of data that GPT-3 is based on.
There are other purposes and use cases that fall between Open AI and Personal AI, depicted in the image below.
X axis: open AI to personal AI. Y axis: low cost to high cost.
Left quadrant shows open ai and top right quadrant shows personal AI
We wanted to close out the year-in-review of Personal AI by giving you a side-by-side comparison of your personal AI model with that of ChatGPT’s.
Open AI is established 2015, Open AI Inc.
Personal AI is established 2020, Human AI Labs Inc.
Open AI envisions a path to Artificial General Intelligence (AGI).
Personal AI envisions a path to Artificial Personal Intelligence (API).
OpenAI’s mission is to ensure that artificial general intelligence (AGI) — highly autonomous systems that outperform humans at most economically valuable work — benefit all of humanity.
Personal AI’s mission is to foster deeper discussions and greater access among humans by creating a personal AI model that remembers and recalls their knowledge and experiences as told by individuals.
Open AI’s language model is called GPT-3, “Generative Pre-training Transformer 3”, a large language model (LLM) trained on internet data. GPT-3 is based on GPT architecture.
Personal AI’s model is called GGT-1, “Generative Grounded Transformer 1”, a personal language model (PLM) trained on personal data. GGT-1 is also based on GPT architecture.
Open AI’s ChatGPT is a variant of GPT-3 that has been explicitly fine-tuned for chat-based applications, such as messaging and chatbots.
Personal AI’s ChatGGT is designed to be a native messaging application with a chat-like interface for human-to-human and human-to-machine interactions.
Open AI generates human-like text by predicting the next word in a sequence based on the context provided by the words that come before it within the scope of an entire internet of data.
Personal AI generates human-like text by predicting the next word in a sequence based on the context provided by the words that come before it within the scope of personal data (intranet).
Open AI features summarization, sentence completion, and question answering within the scope of the entire internet of data.
Personal AI features summarization, sentence completion, and question-answering within the scope of personal data.
Open AI’s Dall-E is an artificial intelligence program that creates images from textual descriptions.
Personal AI memory stack uses a memory block data structure that encapsulates text and image data but does not yet have personal image generation capabilities.
Open AI’s GPT-3 model is about 170B parameters and scales vertically on 10k+ GPU clusters. GPT-4 is projected to be 170T parameters.
Personal AI’s GGT-1 model is about 120M parameters, scales horizontally with GPUs, and is designed to run on the edge, such as a mobile device.
Open AI output loses data attribution and ownership due to data aggregation and anonymization.
Personal AI output preserves data attribution and ownership rights using web3 technology.
Open AI’s GPT-3 model is pre-trained a few times per year, given the nature of large models. Because of this, the model is unsuitable in cases where the recency of the data is critical.
Personal AI’s GGT-1 model is continuously trained overnight, given the micro nature of the models. Because of this, the model is more suitable in cases where day-to-day personal data is critical.
Open AI use cases are most optimal for business-to-consumer engagement applications such as customer service chatbots, social media assistants, marketing content, and as a tool for researchers and writers.
Personal AI use cases are most optimal for human-to-human engagement and access applications such as deeper 1:1 chats, draft responses to text messages, emails, personal knowledge, synthesis-driven debates, and etc.
Open AI owns the data and the models because the company aggregates and structures the data and trains the models for businesses and consumers to use.
Personal AI users own the data and the models because individuals use the no-code Personal AI platform to create their structured memory stack and self-serve their personal AI model training to use and communicate with other humans.
Open AI’s GPT-3 based applications help you create from the internet.
Personal AI’s GGT-1 based experiences help you connect with humans.
Stay tuned for the Personal AI 2.0 update in the new year as we double down on the “personal” in Personal AI.
Updated 4 days ago