"stochastic parrot" is an interesting shibboleth for a certain school of thought on the capabilities of LLMs, but i think it is uncharitable to make the leap straight to "has no clue of how neural networks work". on some level, all of us who spend our free time enough to know what "stochastic parrot" refers to have some idea how NNs work, and on another, none of us know how NNs really work.
we could all do with a bit more humility dealing with this topic and each other.
That's fair, but it is a negative take that disregards all emergent properties. If you strip all emergent properties from it, there is nothing left. The same thing is true of all biological systems.
Why bother being "human". We are all just a bunch of cells exchanging chemicals and electrical signals. That's all there is to it. There is no reasoning, just a bunch of signals going back and forth.
How honestly can you say that "emergent properties" are real if you haven't really seen the training data or you don't actually know how the thing works?.
It stand's to reason that the bigger the model, the more likely you'll get an answer to a question you're looking for. Even the apparently "tricky questions".
Even things like translating code from one coding language to another...
Anyway, maybe we are ALL stochastic parrots (including ChatGPT-10) and that's all we'll ever be...bravo.
Emergent properties are never "real". They just are and you can see them happening, but "underneath" it's nothing.
Edit: I meant to say I don't need access to training. By experimenting with in/outputs you can get a basic picture. I don't need to see biological scans to say something about your personality either.
I do not. What I say is that I perceive them. Their realness is a non-issue (to me). "Factual", you mean by "authorities"? I do get your point, but I think you overthink the issue. If you see something, it is there. It can be illusory, sure, but think about why that matters.
we could all do with a bit more humility dealing with this topic and each other.