I'm surprised more bloggers haven't connected ChatAI (not a trademark) with "pabulum" as a name for the output, a kind of prose we find readable and nutritious, and predictably so, given how it's made in the sausage factory. Or in punk: "AI puke".
The cogsci crowd tends to imagine models float around in neuronal tissue somehow, behind the eyes, whereas the true Markov sheet is the illuminated manuscript, Dark Ages, monks at work. The human training data accumulates in Real Memory, meaning reality, and not redundantly in neural tissue. If we wish to explain photographic memory... not here, not now. Not something most of us have.
When the Library at Alexandria burned, however that went, mainly to keep soldiers warm some people tell me, this was akin to neuronal tissue burning, in the sense of our losing the principal medium of civilizational storage: papyrus or whatever.
But there's this objection that if no one is "thinking about X" then the deeply buried scrolls regarding X must be moribund, and true, they are. Being moribund has nothing to do with being a recording medium for some collective Mind. Temperature matters. A kind of thermodynamic vector equilibrium is still in the picture, at minimum.
Back to pabulum: there's an upside to our having filled the cloud with chatter, in terms of now creating mashup versions of what we've already said, reaping the sewn. That the math is strong enough to find these least action geodesics through a cogspace (game space, namespace) seems miraculous, bravo AI, let me be the first to convey my congratulations.
Or is it RI (real intelligence) we should be applauding? Lets not sell ourselves short. We were inverting matrices long before Nvidia did that for us. I applaud Dr. Hinton and the nD polytopal geometry that was harboring elements of ML, even earlier in the University of Toronto tradition, thanks to Dr. Coxeter.
The Markovian sense in which the present serves as a barrier/blanket twixt the future and past, does not require human experience (bandwidth) explaining the passing down, through time, of these or those memory systems (scrolls, records, files...).
An old and interesting language might reawaken and once again become readable / understandable to a receptive bunch. Receptivity begets transmissivity in many cases.
