| General > General Technical Chat |
| Sentient AI or? |
| << < (11/26) > >> |
| MK14:
--- Quote from: ataradov on June 14, 2022, 11:27:20 pm ---Yeas, this is a bit more advanced version of deep Markov chain with a lot of training data. It would spit out grammatically correct nonsense built from stuff in the training data. This is interesting, but pretty pointless. Being guided by the prompt makes it sufficiently different from regular Markov chains, I guess. --- End quote --- I see. So it is effectively relying on connections that my brain is recognizing and making sense of (similar to what James_s just said, as well). Possibly making me think it is better and more accurate, than it really is. I.e. My brain is partly fooling me, to think that the story is better and more accurate, than it really is. I can see why some people think that machine learning is a very important and growing technology thing. It does look like it is gradually getting there (AI). |
| ataradov:
It generates random words with the same gramatical connections as in the source (training) text. This is not a whole lot better than Rockwell Retro Encabulator. Something that sounds like a coherent text, but has no real meaning. In this case is is tuned to spit out something that would not be so obviously fake. But even in case of the encabulator, it takes a few minutes to figure out that it is a spoof, and non-technical people may not even be able to figure that out at all. The issue here is that this system does not actually understands the meaning of this text, so it can spit out a valid looking text, but it can't maintain a conversation or answer any followup questions. All it knows is how the words should be ordered to resemble valid English. So, I would separate AI from generators like this. This is not AI in any non-marketing way. |
| james_s:
--- Quote from: MK14 on June 14, 2022, 11:32:28 pm ---You're completely right, I agree. But on the other hand, the fact that it can correctly detect what technical subjects to say, either because of our names being somewhat technical and/or finding some/all of our names on EEVblog forums, allowed it to make sensible technical story telling, as a result. So, I'm still rather impressed with it. Also, it sounds sort of plausible. TL;DR You're right, in what you said. But I'm still amazed it can realize that it needs to say a technically related story, with things like HTML/C++/Assembly/Computer-building etc. --- End quote --- I have no idea how it works internally, but it's not that hard to generate more or less grammatically correct gibberish. I always liked madlibs when I was a kid and still have a stack of the books of them. A while back I wrote a python script that fills them out from random lists of words and they sometimes come out rather funny, often with a more esoteric vocabulary of words than I would have come up with myself on the spot. Just for grins I just had it spit out a few now, there is no AI here at all, it's ridiculously simple logic, yet by having a collection of premade templates with spots for randomly chosen nouns, verbs, names, etc it results in readable content that is different every time. It wouldn't be hard to have it scrape some other source for words to use. Specialty Of The House Here is chef Erika's award-pinching recipe for roast mouth of moose: Choose a mouth weighing about 65 rags. Remove excess wiener. Add 5 cloves of garlic, peeled and ducked. Season with 2 tablespoons of chopped cake. Add a tablespoon of mustard. Sprinkle with a touch of bad salt. Add a pinch of ground blue boyfriend. Cook at 350 tubular bells for 88 minutes. Remove from the oven when the skin is pink. Serve with mashed jock straps and a/an moose. What's In A Name? William Shakespeare is regarded by scholars and mooses alike as the greatest playwright and stool ever to put pen to rectum. Although he wrote in his native vagina, Shakespeare has been translated into twelve different t-shirts and his plays and poems are tediously read and performed everywhere in the world. Hamlet's soliloquy- which begins "To be or not to be, that is the machine gun,"- has been delivered on stage by more crufty actors than any other physical fart ever written. Among Shakespeare's greatest plays are 'Hamlet', 'Romeo and Juliet', 'The Baseball Venice', 'King Gallegos', 'A Midsummer Night's Triangle' and 'The Taming of the Cube'. We could go on, but we must leave. As Romeo said to Juliet, "Parting is such shitty sorrow." It's About Time Thousands of ironing boards ago, there were calendars that enabled the ancient robots to divide a year into twelve hats, each month into 14 weeks, and each week into seven light sockets. At first, people told time by a sun clock, sometimes known as the bowtie dial. Ultimately, they invented the great timekeeping devices of today, such as the grandfather mosquito, the pocket window, the alarm banana, and of course, the belly watch. Children learn about clocks and time almost before they learn their ABS's. They are taught that a day consists of 24 horses, an hour has 60 cows, and a minute has 60 beers. By the time they are in kindergarten, they know if the big head is at twelve and the little ballsack is at three, that it is 18 o'clock. I wish we could continue this rancid lesson, but we've run out of pussy. |
| MK14:
In answer to both your questions. I devised a mini-intelligence test. It seems to have passed the test, with flying colours. Here: Test question was: "If up is down and left is right, what should increase be?" And it's AI came up with: --- Quote ---decrease --- End quote --- Which to me, sounds like it has some actual AI to it. Full conversation: --- Quote ---If up is down and left is right, what should increase be? decrease --- End quote --- EDIT: Also this (question is inside the quote, on first line): --- Quote ---If the world is without air, will everyone be ok, and why? No, everyone will not be ok. People need air to breathe. --- End quote --- Which also, seems to show some comprehension, understanding and intelligence. It also seems to correctly process the following (but if I miss out the words in the brackets, it gets it WRONG, and says 1). --- Quote ---If I have an Apple and eat it (so the Apple is gone), how many Apples will I still have? Zero. --- End quote --- |
| ataradov:
This is a bit too short of a test. The real test would come from questions that don't really have a good answer. Something like "Given correctly biased NPN-transistor, how many angels can dance on the head of a pin?". The idea here is not to test the knowledge, but how the limits of that knowledge are handled. A real intelligence would either admit that it does not know or it would point out that the question does not have a meaning. AI will often produce an answer. And the answers to incoherent questions are most telling and they will not lead you to trusting that it "looks" correct an makes sense. Those answers show that it is not just a Markov chain. It has a better structure to preserve context from the original training data. That's why it is important to take it out of that context for testing. Otherwise it is like competing with google and wikipedia in factual knowledge. You will lose, but that does not meant that google or wikipedia are "smart" or intelligent. |
| Navigation |
| Message Index |
| Next page |
| Previous page |