Context is intuitively important, but people rarely put themselves in the LLM's shoes.
What would be eye-opening would be to create an LLM test system that periodically sends a turn to a human instead of the model. Would you do better than the LLM? What tools would you call at that moment, given only that context and no other knowledge? The way many of these systems are constructed, I'd wager it would be difficult for a human.
The agent can't decide what is safe to delete from memory because it's a sort of bystander at that moment. Someone else made the list it received, and someone else will get the list it writes. The logic that went into why the notes exist is lost. LLMs are living the Christopher Nolan film Memento.
The canonical example I use is how good are (philosophical) you at programming on a whiteboard given one shot and no tools? Vs at your computer given access to everything? So judging LLMs on that rubric seems as dumb as judging humans by that rubric.
I looked at Screenpipe but it doesn't fill me with confidence. The ideal solution would be a device-agnostic self-hosted service for archiving and searching screenshots. Would make data migration super simple. And then the app snapping the screenshots could be almost trivial.
It's a bundle of 24 high quality games, by talented, diverse and motivated game developers for ~$8 each. And it comes with a free console. Art is worth paying for.
I feel like this is missing the point of Vim a little bit. It's not about using hjkl instead of arrow keys. It's about progressively learning a vast and useful language specifically made for editing text efficiently. And then realizing you can use every single part of that language in a macro. And then realizing you can use that same language to orchestrate those macros.
The music industry has been fighting against the natural economic theories of capitalism ever since the mp3.
Digital files have infinite supply. Draw a classic supply and demand diagram using that information and look at where the price should be. The supply curve is a vertical line at infinity on the x-axis. The demand curve crosses it at $0. Everyone knows and feels this intrinsically.
The artist is the thing with extremely limited supply. Something like Patreon will end up being the correct model.
This argument falls apart because it is using the incorrect definition of “supply”. The “supply” is the amount available on the market. Just because the cost of duplicating a digital file is nearly zero does not mean that the supply is infinite. The supply is limited by laws that restrict who is allowed to duplicate the file and the limited number of people who are willing to break those laws.
Why do you assume this has been tried? It's not even clear what the game is. In this setting, what state and actions would the algorithm have access to?
In some games it could find an equilibrium where it could keep the game going on indefinitely by moving back and forth, for example (which won't work in a game like Go[1], though).
Context is intuitively important, but people rarely put themselves in the LLM's shoes.
What would be eye-opening would be to create an LLM test system that periodically sends a turn to a human instead of the model. Would you do better than the LLM? What tools would you call at that moment, given only that context and no other knowledge? The way many of these systems are constructed, I'd wager it would be difficult for a human.
The agent can't decide what is safe to delete from memory because it's a sort of bystander at that moment. Someone else made the list it received, and someone else will get the list it writes. The logic that went into why the notes exist is lost. LLMs are living the Christopher Nolan film Memento.