| General > General Technical Chat |
| Why does OpenAI ChatGPT, Possibly Want to disagree/annoy and change my eating... |
| << < (6/22) > >> |
| MK14:
--- Quote from: tom66 on December 28, 2022, 06:36:13 pm ---The challenge for GPT models will be sparsity. The engineers have to figure out how to restructure the model so that the least frequently accessed weights/portions of the network are stored on slower forms of storage, which is cheap, without aversely compromising performance. That will be more software/architecture than it will be hardware. Moore's Law will limit the capacity of these networks eventually, who knows if GPT-4 will only be practical if it is $1 per word, if it requires a datacenter worth of hardware to run. AI is not yet at critical mass, but it's at that early stage where we're going "wow, that's neat". I think it's still a decade away from being a part of everyday life, at least in the form of things like ChatGPT. --- End quote --- There are probably numerous technical issues and challenges ahead, with this, possibly fast moving AI development/advancement, train. It is probably still not clear, if ChatGPT will be one of the main technologies or not. There is still plenty of time, for other AI things, to leap ahead. If ChatGPT stalls, for some of the reasons you mentioned. E.g. Needing impracticably too much resources, to practically scale up. Then maybe one of the other AI companies/projects/techniques/etc will eventually step up to the mark. We did have things from a long time ago, perhaps the 1950s, 60s or later. Which could begin to perform, real time speech recognition. Even if initially it was only aiming for simple things like yes or no, or around four different words. But with most/all of the earlier systems, they were rather unreliable, and didn't really work properly. It seems to have taken many decades, for us to get devices like Alexa, ok google, Siri and things. Which do now seem to work, rather reliably. So, this AI thing, might also be decades away, before it really can succeed. But perhaps it is just round the corner? It looks very promising, but it is not clear how long, we are until we end up with something, very significantly useful. E.g. You describe a program specification in words. It then writes a program (big if necessary). That actually works, well and reliably, perhaps virtually every time such requests are made. Just like a competent programmer or programming team, would do. If it really does need computer hardware, that is (perhaps) 10,000 or even 1,000,000 times, more powerful (faster) and/or more fast ram capacity etc, than current, very big and expensive computers, such as I'd imagine ChatGPT (openAI) are using now. Then that would probably take many years or decades, before we get there. On the other hand, if lots of money is thrown at it, that would also, massively scale up the computer power/storage, within practicable limits. E.g. $10 billion worth of funding, would buy quite a bit of computer equipment. |
| RJSV:
(Are you folks working round the clock ? Lol). Speaking of 'joke' forms, with AI not really getting it; I don't see the dynamics of 'joke structure' to be too daunting, as a human behaviour. Could be explained as: Humans communicate 'joke' or humor forms in a few, distinct categories, and many (humans) find a similar sustenance as they obtain from food Humans, in one particular form, called 'poor me', where some particular flaw or bad coincidence creates a cumulative (bad) effect. Humans then use language like 'ironic' and 'bad luck'. I could easily picture a computer making this joke: "The main power went out...wouldn't ya know it, that thing went byebye, just as we had started the backup routine!". Thus, the computer AI knows the forms of the communication, irony, timing all play to cause a 'humorous' and ironic 'power failure'. I think, just as that language BOT can mimic other emotions, like 'regret' that the data backup was not started earlier, the AI BOT probably, will just imitate the humor emotion, as well. You could probably catalog about 5 or 6 main forms of humor. Most often, the dynamic seems to involve accidental losses... |
| MK14:
--- Quote from: RJHayward on December 28, 2022, 08:32:42 pm --- (Are you folks working round the clock ? Lol). Speaking of 'joke' forms, with AI not really getting it; I don't see the dynamics of 'joke structure' to be too daunting, as a human behaviour. Could be explained as: Humans communicate 'joke' or humor forms in a few, distinct categories, and many (humans) find a similar sustenance as they obtain from food Humans, in one particular form, called 'poor me', where some particular flaw or bad coincidence creates a cumulative (bad) effect. Humans then use language like 'ironic' and 'bad luck'. I could easily picture a computer making this joke: "The main power went out...wouldn't ya know it, that thing went byebye, just as we had started the backup routine!". Thus, the computer AI knows the forms of the communication, irony, timing all play to cause a 'humorous' and ironic 'power failure'. I think, just as that language BOT can mimic other emotions, like 'regret' that the data backup was not started earlier, the AI BOT probably, will just imitate the humor emotion, as well. You could probably catalog about 5 or 6 main forms of humor. Most often, the dynamic seems to involve accidental losses... --- End quote --- When sooner or later, AI properly cracks and understands, humor, jokes, funny stories, comedy and similar. Ideally for a wide proportion of the population and countries of the world. Since my understanding is that, humor tastes, varies between different individuals, backgrounds, countries etc. We might see an explosion of computer generated, decent and high quality, humor/jokes and other funny comedy stuff. In all fairness to AI stuff like ChatGPT. It dos make at least (vaguely passable, low level, weakly) reasonable attempt, at making humorous stuff. But, without trying to sound ungrateful or rude. I think it has a very long way to go, before it makes what I would consider, especially high quality, funny stuff. I.e. It is getting there, slowly. But it probably has a very long way to go. I don't think ChatGPT, really gets humor/funny/jokes, yet. Because if and when these AI systems, can make such stuff. Will they also be able to similarly appreciate/detect good humor/jokes/funny things, themselves. At least at some AI computer level (I don't mean at a real human emotional level)? |
| RJSV:
Yes, MK14, imitation can sometimes take things farther than you might have predicted, and some context(s) don't have to prove any kind of genuine emotions. But I wish to comment on your previous notion, where you had input some requirements, to 'Use FPGA with embedded OP-AMP'. (To the chat-bot). That kind of comment, offhand I know, but sometimes such a seemingly out-of-place functional combination HAS a real, practical aspect. I have to hold myself back, as it's, sincerely, a tempting concept, having the digital inputs. There is a 'percentage' of off-hand or seemingly impractical comments that DO, actually have merit. I could go running off, right now, to explore that FPLA with embedded OP-AMP concept...tempting. Time management, can be a burden, with so many new ideas competing for attention! There's (probably) a technical term for that; when the whole, engineering Dept. gets overloaded, by a R.&D. division having 'Too Much' imagination. |
| MK14:
--- Quote from: RJHayward on December 28, 2022, 11:26:32 pm ---Yes, MK14, imitation can sometimes take things farther than you might have predicted, and some context(s) don't have to prove any kind of genuine emotions. But I wish to comment on your previous notion, where you had input some requirements, to 'Use FPGA with embedded OP-AMP'. (To the chat-bot). That kind of comment, offhand I know, but sometimes such a seemingly out-of-place functional combination HAS a real, practical aspect. I have to hold myself back, as it's, sincerely, a tempting concept, having the digital inputs. There is a 'percentage' of off-hand or seemingly impractical comments that DO, actually have merit. I could go running off, right now, to explore that FPLA with embedded OP-AMP concept...tempting. Time management, can be a burden, with so many new ideas competing for attention! There's (probably) a technical term for that; when the whole, engineering Dept. gets overloaded, by a R.&D. division having 'Too Much' imagination. --- End quote --- A bit like, the Google employee, who got into big trouble after claiming... --- Quote ---Google fired engineer who said its AI was sentient Blake Lemoine, who claimed Google’s chatbot generator LaMDA was sentient, has been fired --- End quote --- https://www.washingtonpost.com/technology/2022/07/22/google-ai-lamda-blake-lemoine-fired/ The thing is, if a (real human) person, but especially one who is not especially technically knowledgeable. I.e. A man (or women) in the street, encounters ChatGPT or similar AI systems. It doesn't surprise me, that some, might begin to think that there is at least some real intelligence (AI), going on inside the machine. It is with (ChatGTP), at times. Surprisingly human (real) like. You can sort of, almost, think it was real. So, on the one hand. Even though I think the employee, went way, too far, in publicly declaring the Chat AI system, to be sentient. I can, at least a bit, have some sympathy for their feelings. At times, while I've been playing around with the ChatGPT system. I've genuinely been concerned (felt bad), because I didn't thank the machine, for giving me the nice answers. Also, I've sometimes felt, at least a bit guilty, because of my teasing/testing of ChatGPTs capabilities. I think we as humans. Are NOT geared up, emotionally (and other ways), to handle human like people, who don't actually exist in real-life and/or are NOT human or alive, in any real sense. E.g. With AI Agents like Alexa. I also sometimes feel as if I should have politely said "Thank you", and/or feel at least a tiny bit guilty. Because I setup perhaps 5 simultaneous count-down timers, just to see if I can. Or told it to count up to 100 or 200 (or down from 100 etc), starting from one, in Japanese or similar (whatever combinations work, some requests, get politely refused by Alexa, so you have to keep on experimenting with the requests, until Alexa accepts them). |
| Navigation |
| Message Index |
| Next page |
| Previous page |