@ferret I think people just use LLMs like a hammer, they have use cases, but for anything that involves math or formal logic, we have far better tools available
@Yogthos I'm more just interested in their functioning. There's nothing in how they work that contains temporality. You can hook them up to software that lets them tell time, but there's nothing in an LLM that isn't measured in terms of token generation, not objective time.
@ferret I mean you could train a neural network to keep a timer, but the way LLMs are currently designed makes them ephemeral by nature, all the context lives externally