[LINK] ChatGPT

[David]

On 4/2/23 22:15, Bernard Robertson-Dunn wrote:
> Now it's getting really confused. The above response is an incoherent mixture of two quite distinct concepts.
>
> A Bravais pendulum swings in a conical manner and is impacted by the earth's rotation depending on if it goes clockwise or counterclockwise.
> From https://www.britannica.com/science/Bravais-lattice
> Bravais lattice, any of 14 possible three-dimensional configurations of points used to describe the orderly arrangement of atoms in a crystal.  [...]

I guess this ChatGPT discussion has just about run its course.  But I think the most interesting aspect is ChatGPT's package architecture and design, and why it sometimes returns such silly responses.

It seems its' most significant component (and probably OpenAI's major I.Property investment) is a novel natural-language model and implementation.  Presumably a front-end component parses a query in an attempt to locate its linguistic elements and uses that to search the 'net, and then a back-end converts the results to natural language again for presentation to the enquirer..

Much of the ChatGPT's charm lies in this language model.  People are impressed by the "intelligence" of an apparently natural, human-like reply, and sounding like the HAL-9000 wouldn't do any harm either.

The language model may well be based on some form of existential calculus (which is distinct from the predicate calculus) - see "Beginning Logic" by E.J. Lemmon (Nelson) and "Word & Object" by WVO Quine if you're a glutton for punishment.

However I suspect that sort of system architecture is prone to "category errors" which are perfectly well formed statements at a purely syntactical level but which make no sense at all: e.g. "the number two is blue".  That example was taken from the Stanford Dictionary of Philosophy at https://plato.stanford.edu/entries/category-mistakes.

All of which, if true, supports a view that the vital ingredient missing from ChatGPT and all "AI" implementations for some unknown time to come is insight, understanding, and ultimately consciousness.  That's not to say AI isn't a very useful tool when used properly, e.g. for a first examination of some medical scans.

But mass unemployment is not imminent IMHO.

Cheers,
David Lochrin