Here he writes about ChatGPT :)
https://mwichary.medium.com/one-hundred-and-thirty-seven-sec...
Here as well: https://electricliterature.com/wp-content/uploads/2017/11/Tr...
And here about Ebooks and Audiobooks:
When I first read this "preface" twenty years ago, the idea seemed implausible to me: How could a system write novels only by being fed other novels, and without simultaneously being a general intelligence? Surely novel writing is AGI-complete!
The machine will employ, as the need arises, the pro- noun "I" and all its grammatical inflections. This, however, is a hoax! The machine will still be closer to a billion chattering parrots—howsoever brilliantly trained the parrots be—than to the simplest, most stupid man. It mimics the behavior of a man on the purely linguistic plane and nothing more. Nothing will amuse such a machine, or surprise it, or confuse it, or alarm it, or distress it, because it is psychologically and individually No One. It is a Voice giving utterance to matters, supplying an- swers to questions; it is a Logic capable of defeating the best chess player; it is—or, rather, it can become—a consummate imitator of everything, an actor, if you will, brought to the pinnacle of perfection, performing any programmed role—but an actor and an imitator that is, within, completely empty. One cannot count on its sympathy, or on its antipathy. It works toward no self-set goal; to a degree eternally beyond the con- ception of any man it "doesn't care," for as a person it simply does not exist.... It is a wondrously efficient combinatorial mechanism, nothing more.
ok but i'm still going to thank chatgpt just in case
- purely symbolic with myriads of symbolic (if-else) clauses similar to expert systems, or
- sub symbolical ANN networks similar to current LLMs
i suspect it was the former in which case that he actually arrived at description of current large NNs is really striking I think
He wrote A LOT about this and explored various consequences, including the simulation argument which he presented in 1960 as a short story "Strange chests of professor Corcoran" - https://przekroj.org/sztuka-opowiesci/dziwne-skrzynie-profes... here's the Polish version).
, he's had extraordinary intuition with this
At this point the definition of AI is practically "Something computers can't do yet", though I'm partial to its corollary "Any sufficiently misunderstood algorithm is AI."
Only because we thought such a robot would be like a AGI servant, not just a single-purpose device like a roomba that can just bounce around the floor.
So, it's not like we've changed our definition of AI (and even less so, AGI). What we did change is what a robot house-cleaner product is (less "C3PO with a broom", and more "single-purpose vacuum cleaner with heuristics to bounce around").
Even for chess playing, when people in the past thought a chess playing machine that would be able to defeat the human champion would have AGI, they did so not because they thought playing chess is enough to signify AGI, but because they thought AGI was necessary to do so.
If someone had explained to them back then that such a future machine would be able to play expert-level chess by mere number crunching of a huge list of moves, and that it wouldn't imply any other thinking facultu, they wouldn't consider that to be AGI.
>At this point the definition of AI is practically "Something computers can't do yet", though I'm partial to its corollary "Any sufficiently misunderstood algorithm is AI.
The practical definition of AI (as used colloqualy, in the market etc) for products is basically "any smart-looking algorithm, with heuristics to do something slightly complex".
It's just that the term is overloaded, and we sometimes say AI when we mean AGI.
I'd argue the problem isn't how the machine does it, but the fact that you've explained how the machine does it. If you explained exactly how a brain worked, people would say "that's not intelligence, that's a just ion circuity."
(Copying from another post): there are neurons in your eye that see something and send signals out laterally so when they (after a specific time-delay) meet another neuron that is currently seeing the thing, they combine to trigger a signal that can be directly interpreted as movement in a specific direction at a specific arc-speed.
Explain each process in the brain like that and it loses its magic. It actually becomes very similar to the debate between free will and meat robots.
Well, even 20 years ago, a simple Markov Chain Tolstoy-based output could very well pass at least for a modernist style novel.
For Solaris (a serious novel) I would recommend the new Kindle translation, if you are reading in English. The old one was a retranslation from French for some reason.
the concept is cool, but the real gem is the fact that it's not really written in the form of a novel. And yet, it's still recognizable as one. I love that Lem explored in bending these definitions
> Trurl realized that a poet's mind is shaped by the civilization that produced him, and that civilization - by the one before it, and so on. So in order to create an artificial poet, he'd have to simulate the whole history of the Universe. Which he did, but then from the results chose only the important bits as inputs for the learning process, otherwise it would take too long...
https://lab.cccb.org/en/did-calvino-dream-of-literary-androi...
> Arguably the oldest known story of something approximating AI can be found in the eighth-century-bc Iliad, Homer’s epic poem of the Trojan War. In it, Hephaestus, disabled god of metalworking, creates golden handmaidens to help him in his forge: “In them is understanding in their hearts, and in them speech and strength, and they know cunning handiwork”.
> ..In her 2018 book Gods and Robots: Myths, Machines, and Ancient Dreams of Technology, Adrienne Mayor describes how ancient cultures explored the idea of artificial life. Ancient Greeks were skilled in metalwork and mechanics and created a great deal of automata, including a puppet theatre that could perform an entire play.
The idea of a poetry-writing machine can also be traced back in history, hinted at in the Kabbalah, or machines of ancient China. (Source?) In the work of Leibniz, "On the Combinatorial Art" (1666), he explores:
> The main idea behind the text is that of an alphabet of human thought, which is attributed to Descartes. All concepts are nothing but combinations of a relatively small number of simple concepts, just as words are combinations of letters.
> All truths may be expressed as appropriate combinations of concepts, which can in turn be decomposed into simple ideas, rendering the analysis much easier. Therefore, this alphabet would provide a logic of invention..
(and i still don't understand how ChatGtp is able to respond to metaprompts like 'be brief', i mean if it is a model that is selecting the next most likely word, then it should not be able to control its own style of output.)
"Be brief" may just be associated with the speech patterns accociated with a shorter way to say stuff. So it's still the same autocomplete, just steered in another direction.
Now i am a bit sceptical of this whole prompting business, still some of it is doing something (otherwise people would not put their work into it)
Also not clear to me how this steering of the automcpleter in another direction is working.
I didn't bother to check the math on this, but in the story there is not enough matter in the universe to in some way encode every possible traditional Chinese poem!
By the way, reading a "double-translation" (Chinese -> English -> Polish) is fascinating--at times it's more than obvious that what you're reading is not what the author has written, but you have no idea at all what the original concept was. (Unlike "single" EN -> PL translations, where I can often figure out the idiom or concept that was used in the original.)
In fact if you formally describe how you generate the permutations that's one of such encodings of these permutations (and the optimal one - see Kolmogorov Complexity :) ).
So there is definitely enough matter to do it. Similarly we can encode PI despite it having infinite number of digits.
Ce-ci n'est-pas une remarque, and all that.
Forget PI. Let's take 1337. Or is it MCCCXXXVII?
Is it 2 or 1.(9)? Or 10 (binary)? Different encodings, same number. Some encodings are just less optimal than others.
Same with text. Is the poem in utf-8 and utf-16 a different poem? What if you zip the file? These are just encodings, and there's no point ignoring the good ones (which for non-random strings are usually programs).
"This is a poem about a haircut! But lofty, nobel, tragic, timeless, full of love, treachery, retribution, quiet heroism in the face of certain doom! Six lines, cleverly rhymed, and every word beginning with the letter "s"!
Seduced, shaggy Samson snored.
She scissored short. Sorely shorn,
Soon shackled slave, Samson sighed,
Silently scheming,
Sightlessly seeking
Some savage, spectacular suicide."I remember reading book about stock exchange Ai predicting terrorist attack casualties and then predicted 100k dead in London
has anyone else read smth like that and remembers the name?
I forgot how much I enjoyed this...