honestly, kind of both. it started as a curiosity. I just wanted to see if it could even work. but after playing with it, I think there’s something deeper.
the most interesting part to me is the idea of apps that self-customize without being told to. they adapt based on how you use them and what you enter, not from settings or configuration. stuff that changes itself in real time, just by being used. we wouldn't build that kind of software today because it's too weird or specific. but with this, it just happens.
still super early and pretty rough, but it feels like a door opened a crack.
But isn't the point of an app being able to maintain a sharable worldview and state?
This "improvised just for you" sounds like a natural fit for a text adventure for example, but I'm still grappling with how this might apply to a real life use case.
One can conceivably imagine imagine LLMs one day transforming states for sharing on-the-fly as well, but it does sound very fragile.
Totally agree. Persistence is a huge part of what we usually mean by state. Nothing stops this from plugging into a real database when needed. But also, nothing stops us from rethinking what a database even is, especially in a world where LLMs have growing context windows and can carry more state internally. Maybe persistence doesn’t always have to mean "write to disk."
In the context of LLMs memory can often mean a broadly defined, application-specific ability to detect, retain and re-use bits of useful context along the journey. In contrast to Persistence which is more like a DB with a schema and clear expectations, reliability, etc.
good question. I think it'll be self-reinforcing, since the context keeps growing with everything the software has done for you, along with every interaction you’ve ever had with it.