Most tech companies aim to make certain parts of their users’ lives easier — Google, for instance, makes it easier to find things on the web, and Meta’s products let users stay connected with friends and family. OpenAI, though, seems to have loftier ambitions.
OpenAI CEO Sam Altman has said that in an ideal scenario, he’d want ChatGPT to remember its users’ entire lives, and then use that data for customization. He described a “platonic ideal” where a powerful AI model possesses comprehensive knowledge of a user’s life, using it to personalize interactions and generate insightful outputs.
“I mean in some sense,” Altman said at the AI Ascent Summit, “I think that platonic ideal state is a very tiny reasoning model with a trillion tokens of context that you put whole life into.”
He envisions a system where the model is “never retrained the weights, never customized, but that thing can reason across your whole context and do it efficiently.”
The scope of this vision is breathtaking. Altman elaborated, suggesting that “every conversation you’ve ever had in your life, every book you’ve ever read, every email you’ve ever read, everything you’ve ever looked at is in there, plus connected all your data from other sources.”
Furthermore, he imagines that “your life just keeps appending to the context, and your company just does the same thing for all your company’s data.”
Altman acknowledged that this is not achievable in the present. “We can’t get there today,” he stated. But he emphasized that he thinks of “anything else as a compromise off that platonic ideal, and that is how I hope we do customization.”
The implications of such a system are profound. Imagine a ChatGPT that not only knows your preferences but also intimately understands your past experiences, relationships, and even your thought processes. The potential for personalized assistance, creative collaboration, and even emotional support is immense. OpenAI already seems to have taken the first steps towards such a system by launching a memory feature, through which ChatGPT can remember past chats.
But there would be many breakthroughs required before such a system could exist to the full extent that Altman envisions. For starters, there are currently no models which can ingest a trillion tokens in their context window. But if context windows get larger in the coming years, and there are mechanisms to collect and store all this user data through their lives, such a system could well become a reality. In theory, a model like this would know every little detail about the user, and would be able to offer unprecedented insights and advice. This collection of life experiences would also be an insurmountable moat if this information couldn’t be ported to another AI company, and users would likely come to rely on such a system to a degree that they would find it hard to operate without it. It’s an ambitious goal, but this is what OpenAI seems to be aiming towards as it looks to build on its early lead in the AI race.