|
Hipster_Doofus posted:Exactly why I thought just going to mandarin and back would be pretty mangled. Otoh I've done some translations to Spanish and back that came out hosed up beyond recognition. English and Mandarin aren't actually as different as you might think. Mandarin grammar is fairly straightforward and has similarities to English. Plus there is no modification of words whatsoever, no verb conjugation or anything, which simplifies things.
|
#
?
Jun 14, 2019 22:34
|
|
|
|
| # ? Apr 27, 2024 21:39 |
|
|
AlBorlantern Corps posted:It isn't though, because to learn top drive a car also requires you to develop fine motor control, and process and prioritize what you see and hear, and ability to think and perceive in 3D space. It's just those are tasks you've already mastered by that age For full use of language, you need to understand what people think, which probably means you have human-level AI. I'm pretty sure we'll have self-driving cars before we have that.
|
|
#
?
Jun 14, 2019 22:35
|
|
|
Cingulate posted:Arguably we have particular machinery specifically for acquiring language (this goes against my own professional opinion, but it's a fairly prevalent opinion in the field), making the difference even more substantial. That's just it. Recognizing objects and navigating among them is a massively complex task brains and eyes were highly specialized to do long before anything like a human language existed. That's been a hard thing to computerize.
|
|
#
?
Jun 14, 2019 22:35
|
|
|
Killer robot posted:That's just it. Recognizing objects and navigating among them is a massively complex task brains and eyes were highly specialized to do long before anything like a human language existed. That's been a hard thing to computerize. Generally, we have breakthroughs in computer vision a good bit before breakthroughs in language. For example, the recent breakthroughs in language modelling have come from taking computer vision-inspired networks (convnets with attention) and applying them to language. And there are specific tasks where machines beat humans in the field of vision; but not really any in language.
|
|
#
?
Jun 14, 2019 22:37
|
|
|
Grand Fromage posted:English and Mandarin aren't actually as different as you might think. Mandarin grammar is fairly straightforward and has similarities to English. Plus there is no modification of words whatsoever, no verb conjugation or anything, which simplifies things. Yep. When I took Mandarin in college we had to give oral presentations with PowerPoints of the English translation. With rare exception we could write it such that we could basically just read the PowerPoints substituting Chinese words for the English. Bring able to bend word order to our will was really handy.
|
|
#
?
Jun 14, 2019 22:41
|
|
|
Cingulate posted:... than language? I'm pretty sure we'll have cars way before real language understanding. Avis mandates a person at the controls at all times and recommends you get complete coverage.
|
|
#
?
Jun 14, 2019 22:43
|
|
|
Cingulate posted:For full use of language, you need to understand what people think, which probably means you have human-level AI. I'm pretty sure we'll have self-driving cars before we have that. Gonna have to disagree with you there boss. Language is a lossy compression algorithm, but it's not necessary to fully model the thoughts of the speaker to decompress it, just to grok the context.
|
|
#
?
Jun 14, 2019 22:48
|
|
|
The Orville: Google Translate Edition
|
|
#
?
Jun 14, 2019 22:49
|
|
|
Cingulate posted:Arguably we have particular machinery specifically for acquiring language (this goes against my own professional opinion, but it's a fairly prevalent opinion in the field), making the difference even more substantial. chomsky can eat a turd
|
|
#
?
Jun 14, 2019 22:57
|
|
|
Cingulate posted:A computer doesn't need to develop fine motor control to drive a car; you also need to develop fine motor control for speaking (tongue placement is extremely delicate!); and understanding how other people tick is probably much more complicated than 3D space. For example, many fairly stupid animals are really good at 3D space. Computers are already a lot better at controlling a car than a human ever could be. The problem is computers are loving stupid and have no real idea what's happening around them. Just feeding them piles of data hasn't worked.
|
|
#
?
Jun 15, 2019 00:07
|
|
|
and that's different from humans how?
|
|
#
?
Jun 15, 2019 00:12
|
|
|
Cingulate posted:Speaking of which, the ubiquity of manually piloted spaceships in Scifi is probably appear as a massive anachronism soon. In Starship Troopers, Ibanez plots a new course that the computer can recognize as better, but the computer can’t plot the course on its own. That always bugged me.
|
|
#
?
Jun 15, 2019 00:36
|
|
|
Son of Sam-I-Am posted:and that's different from humans how? See previous statement, eg: fire trucks
|
|
#
?
Jun 15, 2019 01:30
|
|
|
Peachfart posted:See previous statement, eg: fire trucks Ladies and gentlemen, I present Exhibit A.
|
|
#
?
Jun 15, 2019 01:38
|
|
|
Zurui posted:chomsky can eat a turd seconded
|
|
#
?
Jun 15, 2019 03:55
|
|
|
Grand Fromage posted:I kind of suspect where translation is now is as good as it's going to get for a while, but I'm one of those people who thinks machine learning/AI is being really overhyped right now and is going to be a much harder problem than a lot of people have been claiming lately. Those commercials where they smugly gush over Microsoft "AI" drive me nuts. ITS NOT REAL AI. We are nowhere near true artificial intelligence, and when it happens it will be one of the biggest events in human history and be all over the news. All they have now are really good computers. Calling them AI is like calling those two wheeled electric balancing boards "hoverboards." 
|
|
#
?
Jun 15, 2019 08:36
|
|
|
Astroman posted:Those commercials where they smugly gush over Microsoft "AI" drive me nuts. ITS NOT REAL AI. Son of Sam-I-Am posted:and that's different from humans how?
|
|
#
?
Jun 15, 2019 08:48
|
|
|
It was snarky and rhetorical you guys.
|
|
#
?
Jun 15, 2019 12:44
|
|
|
A significant problem with artificial intelligence, whether you're talking about self-driving cars or machine translation or anything else, is that the amount of effort that it takes to go from 95%-100% complete can often be hugely more than it took to get to 95% in the first place. In some domains, this is acceptable. In others, it is not. If a machine translation gets 95% correct, that's probably okay for a lot of contexts but you'd not want to use it for translating contracts, court testimonies, or anything else where small changes in meaning can have a huge effect. For a self-driving car, a car which wrecks itself even once a year on average is still unacceptably bad, for obvious reasons. Krispy Wafer posted:In Starship Troopers, Ibanez plots a new course that the computer can recognize as better, but the computer can’t plot the course on its own. The complexity of coming up with a new course, and the complexity of verifying that a given course is better than an earlier option, are not comparable, so I will rate this "plausible."
|
|
#
?
Jun 15, 2019 15:49
|
|
|
Computers are very good at tasks where all the parameters are defined, and very bad at tasks where parameters are unknown. Humans are mediocre at the former and mediocre at the latter. Mediocre is still better than very bad though.
|
|
#
?
Jun 15, 2019 16:09
|
|
|
PT6A posted:
In fact the NP complexity class of problems is entirely* defined as "we can quickly verify provided solution, but finding it is hard". * At least when translated to lay terms
|
|
#
?
Jun 15, 2019 17:05
|
|
|
Isn't another problem of language vs. self driving is that the structure of a lot of languages is in place. Noun, verb, adverb, adjective and where they go is like 85 percent of the battle. Driving on the road is not that. Sure, you have to drive on a road but there is not like structure around it? does that make sense?
|
|
#
?
Jun 15, 2019 17:20
|
|
|
Krispy Wafer posted:In Starship Troopers, Ibanez plots a new course that the computer can recognize as better, but the computer can’t plot the course on its own. I don't know the scene, but in principle, this isn't unreasonable. Many problems are hard to solve but easy to verify. Pathing problems are often like this (NP-complete in CS jargon).
|
|
#
?
Jun 15, 2019 18:02
|
|
|
Astroman posted:All they have now are really good computers. To be fair, strong AI would also just be a really, really good computer Mooseontheloose posted:Isn't another problem of language vs. self driving is that the structure of a lot of languages is in place. Noun, verb, adverb, adjective and where they go is like 85 percent of the battle. Driving on the road is not that. Sure, you have to drive on a road but there is not like structure around it? Not to me It doesn't help that language is handled by a part of our brain other than the conscious, thinking part--meaning we don't have direct access to peek at our brain's rules for How to Speak English. We can query it for whether a given sentence is grammatical or not, but that's it--we have to try to build up the formal rules just from those observations. And that's hard it turns out, much much harder than anybody who hasn't studied linguistics would ever suspect.
|
|
#
?
Jun 15, 2019 18:22
|
|
|
That's absolutely right. We don't know how we speak and read and write. How do you make a computer that can? For that matter we don't know how we think...
|
|
#
?
Jun 15, 2019 18:32
|
|
|
Hipster_Doofus posted:That's absolutely right. We don't know how we speak and read and write. How do you make a computer that can? For that matter we don't know how we think... Psch, I know how I think and speak and write. Badly.
|
|
#
?
Jun 15, 2019 18:42
|
|
|
Hipster_Doofus posted:That's absolutely right. We don't know how we speak and read and write. How do you make a computer that can? For that matter we don't know how we think... Ehh potentially several ways even so, but we can't intentionally make one that does it the same way we do, so they will always be truly alien
|
|
#
?
Jun 15, 2019 18:43
|
|
|
What does any of this have to do with 500 cigarettes?
|
|
#
?
Jun 15, 2019 20:30
|
|
|
The Bloop posted:Ehh potentially several ways even so, but we can't intentionally make one that does it the same way we do, so they will always be truly alien Fred Jelinek posted:Every time I fire a linguist, the performance of our speech recognition system goes up. I.e., trying to model our language AIs to work like we think human language works, is a dumb idea. So we just use a lot of data and powerful semi-general architectures instead. But interestingly, they recover much of what we would engineer into them intentionally all by themselves: https://arxiv.org/abs/1905.05950 Cojawfee posted:Computers are very good at tasks where all the parameters are defined, and very bad at tasks where parameters are unknown. Humans are mediocre at the former and mediocre at the latter. Mediocre is still better than very bad though. For example, current speech generation systems are already better than many Star Trek computer voices/Majel Barrettbeing instructed to talk monotonically.
|
|
#
?
Jun 17, 2019 09:29
|
|
|
PT6A posted:The complexity of coming up with a new course, and the complexity of verifying that a given course is better than an earlier option, are not comparable, so I will rate this "plausible." They can do it today with GPS. Granted there’s generally only a handful of possible routes for our pocket computers to calculate, but by the time faster than light travel is commonplace I’d hope our computers could handle the extra CPU cycles. Not to mention the initial route is going to be recalculated and modified continuously for the duration of the trip. And I realize the absurdity of asking for realism from Starship Troopers.
|
|
#
?
Jun 17, 2019 09:56
|
|
|
I'm not sure I buy that the Starship Troopers scene is that plausible. The thing with space (especially interstellar space) is that it's very big, very empty, and celestial bodies tend to move in very predictable ways. Barring some unforeseen thing (like that asteroid they almost crash into), it should be a trivial task to calculate the optimal course between two points in space, especially if all you're interested in is going between two star systems like what happens in Starship Troopers.
|
|
#
?
Jun 17, 2019 13:41
|
|
|
We don’t know how their FTL works. It might be like Star Wars where you have to account for mass shadows and plot around them. Much like most sci-fi it shouldn’t really be thought about too much and just taken as a way to say ‘she’s good.’
|
|
#
?
Jun 17, 2019 14:44
|
|
|
I think you all are missing the point which is obviously how rad those futuristic nineties computer readouts look.
|
|
#
?
Jun 17, 2019 16:34
|
|
|
Crew hard at work on S3!
|
|
#
?
Jun 28, 2019 19:00
|
|
|
Seth is in "The loudest voice", the Roger Ailes/Fox miniseries on Showtime. Ballsy.
|
|
#
?
Jul 9, 2019 13:38
|
|
|
https://twitter.com/orvillegame/status/1148638915391295493?s=21
|
|
#
?
Jul 10, 2019 05:29
|
|
|
The question now is if Fox approves.
|
|
#
?
Jul 10, 2019 10:42
|
|
|
Fans make interactive starship video game: Star Trek-shuts that poo poo down immediately even thought he creators didn't want to make any money off it Orville-raves about it and makes it official and probably pays creators well This is what happens when shows are done BY FANS.
|
|
#
?
Jul 10, 2019 12:52
|
|
|
Mokinokaro posted:The question now is if Fox approves. Seth MacFarlane retweeted the project a few times, it's pretty unlikely there will be any action.
|
|
#
?
Jul 10, 2019 22:09
|
|
|
|
| # ? Apr 27, 2024 21:39 |
|
|
Will it support VR?
|
|
#
?
Jul 11, 2019 01:53
|
|
