There's no inherent, technological reason to feed the historic data into the prompt, though. You could train a network with a second input for model-states, and during inference, feed into that input historic model-states that occured for the same LLM when consuming input that is somehow associatively related to whatever input is currently being consumed. Can you definitely show that this would be meaningfully different from how humans "experience" memories?
There is absolutely ZERO evidence or possibility that LLMs are currently conscious, everyone working in AI knows this, (and don’t listen to the marketing hypes).
LLMs still can’t reason (don’t confuse with the marketing word) , plan, or understand the world, no emotion, no subjective experience, no memory, no intent.
And yes I would change my opinion if it were logically different, other architectures may pave the way there, but not LLMs
There is no "evidence" that anything has consciousness, nor could there be. To clarify, by that I don't mean nothing is conscious, I mean that it's never been an evidence-based proposition.
We are conscious because we declare ourselves to be, by fiat, not because of evidence or a well-considered rationale. Our conception of consciousness or understanding isn't formal, checkable or objective, it's entirely grounded in subjectivity. As such, any and all arguments for or against consciousness or understanding in LLMs are based on feelings. We feel that we are conscious, therefore we are. Regarding anything else, "we'll know it when we see it." There's nothing rigorous about it.
Yes there is no definitive proof, but the reverse is also not true. Please read some science on that in neurology and biology. Yes it is hard to define, though there are definitions and those are not fulfilled by a LLM, What’s missing is a bearer: no state persists between calls, nothing is maintained against pressure, nothing costs the system anything.
though more an more animals seems to be conscious as we previously thought
What evidence would take you from "LLMs aren't conscious," to "LLMs are definitely conscious"? Just asking because the responses tend to be genuinely interesting and give insight into what's important to people.
Memory is definitionally the faculty of the mind by which data or information is encoded, stored, and retrieved when needed. There's no special carve out for 'except when it's an LLM.'
Why, exactly, is this distinction relevant? The brain must store episodic memory somewhere too. From a cursory read, it seems to me that our experiences are processed through the hippocampus and amygdala into a high-level annotated encoding of historical data, stored somewhere else (unclear), fetched based on various triggers, replayed and reinterpreted by the neocortex, and, I assume, sent back through the hippocampus once more for updating. Sure, it's all in the brain, but I don't think physically separating the storage makes it fundamentally different.
I mean, I've replayed old memories in my head before. Sound, image and all. That's kind of like sending historic data into a prompt, isn't it? Of course it's not exactly the same thing, but the difference isn't exactly load-bearing, if you'll excuse the claudism. I feel like we're grasping at straws.
Actually not quite true: Memory is distributed across synaptic weights rather than sitting at an address or location.
The difference is that human retrieval isn’t a read operation, reconsolidation means recalling a memory destabilizes it and re-stores an altered version. The act of remembering modifies the thing doing the remembering. An LLM’s weights don’t change when you put history in the prompt, and the prompt is discarded intact afterward. There is no state that would indicate any form of consciousness.
There is no form of subjective experience.
We used analogies from nature to describe computers and the same vocabulary and now because of this people now draw reverse conclusions that are not necessarily true