6 ms·

It sounds nice but it's just one non sequitur after another. A very sleek looking plane that is incapable of flight. I expect the exact opposite from what I wo

It sounds nice but it's just one non sequitur after another. A very sleek looking plane that is incapable of flight. I expect the exact opposite from what I would qualify as AI: something that doesn't look like any plane a human would design but yet it flies! Continuing on that parallel: it took ~400 years from the flying machines like the one designed by Leonardo da Vinci to transition from a superficial imitation of birds to a design that didn't look like birds but could actually fly.

aaron695 6y ago
It's Blackhat AI. Designed to confuse humans. Compared to chess AI for instance which I think still can't play like humans.
- orwin 6y ago
  Did you see Alphazero vs stockfish chess matches? I looked into 7 of them, and of those 7, 3 of them could have been played by a human (2 victories one draw). It was impressive.
- ttsda 6y ago
  Chess AI can play like humans, and win consistently (stockfish), and it can also play in completely novel ways, and win even more (alphazero)
  - SurfingToad 6y ago
    Stockfish doesn't play like humans. That's how sites like Lichess detect cheating; you can tell Stockfish moves apart from human moves. AlphaZero/Leela Zero, on the other hand, plays like humans.
m12k 6y ago
I mean, it's an amalgam of text written by humans - it's not even trying to make a plane that flies, it's trying to make a plane that looks like all the planes designed by humans - and largely succeeding. I'm sorry that's not a task you consider interesting, but the fact that a machine learning model trained in an unsupervised fashion can succeed that much at anything non-trivial is mind-boggling to me, and makes me hopeful that other tasks will also be within reach relatively soon.
- logifail 6y ago
  > trying to make a plane that looks like all the planes designed by humans [..] I'm sorry that's not a task you consider interesting Trying to create text that looks like text written by humans? Is that really what GPT-3 is designed to do? I'm reminded of one of my wife's responses, on being told that a person speaks N languages fluently ... "Does s/he have anything interesting to say in any of them?"
  - segfaultbuserr 6y ago
    >> trying to make a plane that looks like all the planes designed by humans [..] I'm sorry that's not a task you consider interesting > Trying to create text that looks like text written by humans? Is that really what GPT-3 is designed to do? My understanding is, it's designed to attempt to "understand" the inner structure of text written by humans, and to create new text using the same structure based on its "understanding". "Understanding" is only defined as the creation of a probabilistic model - it doesn't "understand" any meaning, only the structure of sentences and paragraphs, Yet, such a mindless machinary can demonstrate a wide range of interesting behaviors. Using the plane analogy: If you feed the airplanes designed by humans throughout history into GPT-3, GPT-3 is able to produce an original design of convincingly-looking airplanes of all nationalities and types. For example, if you say "1960, turboprop, Britain", it gives you one, and it's also capable of mixing things, such as putting a jet engine on a biplane, or designing a hypothetical WW2 Allies plane using Axis military technology. If you are lucky, some of the planes come with semi-functional systems. Some might be able to ignite its engine before it explodes, and few are even capable of taking off before it explodes (and some are unable to take off, but if you put them in the mid-air, it may be able to travel for one minute before it explodes). And it's all done under unsupervised learning, and without any domain-specific knowledge on Newtonian mechanics, aerodynamics or aviation. All it needs is an astronomical number of blueprints collection. To me it's an impressive achievement, even if it doesn't understand how to design airplanes.
    - m12k 6y ago
      I read somewhere that unlike GPT-2, GPT-3 "understands" addition. So if you feed it "74 + 129 = " it will come up with 203, even if that specific math problem never occurred in any of the text it was trained on. That to me is the most impressive result so far - not only is it possible, but also more efficient, for the model to actually encode the rules of addition instead of just memorizing the results for specific math problems. This to me seems like the dawn of insight and understanding, and I'm super excited to see what comes next.
lifeformed 6y ago
Well, what are you expecting? An actual explanation for the grand theory of everything? Or just a fun story as a response to the prompt? To that, it does a decent job, and that was all expected of it. Give a human the same prompt and see what they say.
leto_ii 6y ago
> It sounds nice but it's just one non sequitur after another. A very sleek looking plane that is incapable of flight. I like this analogy, especially because it alludes to the Dijkstra quote [1]. In this light it's probably unfruitful to try to create submarines that mimic swimming or that look like fish - you'll just end up with fancy looking stones. [1] "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim"
- rrrrrrrrrrrryan 6y ago
  Yes - in English, the verbs "swim" and "think" both imply that the action is being performed by a lifeform. "Fly," for some reason, does not. I wonder if this holds up in other languages as well.

aaron695 6y ago

It's Blackhat AI. Designed to confuse humans. Compared to chess AI for instance which I think still can't play like humans.

orwin 6y ago
Did you see Alphazero vs stockfish chess matches? I looked into 7 of them, and of those 7, 3 of them could have been played by a human (2 victories one draw). It was impressive.
ttsda 6y ago
Chess AI can play like humans, and win consistently (stockfish), and it can also play in completely novel ways, and win even more (alphazero)
- SurfingToad 6y ago
  Stockfish doesn't play like humans. That's how sites like Lichess detect cheating; you can tell Stockfish moves apart from human moves. AlphaZero/Leela Zero, on the other hand, plays like humans.

m12k 6y ago

I mean, it's an amalgam of text written by humans - it's not even trying to make a plane that flies, it's trying to make a plane that looks like all the planes designed by humans - and largely succeeding. I'm sorry that's not a task you consider interesting, but the fact that a machine learning model trained in an unsupervised fashion can succeed that much at anything non-trivial is mind-boggling to me, and makes me hopeful that other tasks will also be within reach relatively soon.

logifail 6y ago
> trying to make a plane that looks like all the planes designed by humans [..] I'm sorry that's not a task you consider interesting Trying to create text that looks like text written by humans? Is that really what GPT-3 is designed to do? I'm reminded of one of my wife's responses, on being told that a person speaks N languages fluently ... "Does s/he have anything interesting to say in any of them?"
- segfaultbuserr 6y ago
  >> trying to make a plane that looks like all the planes designed by humans [..] I'm sorry that's not a task you consider interesting > Trying to create text that looks like text written by humans? Is that really what GPT-3 is designed to do? My understanding is, it's designed to attempt to "understand" the inner structure of text written by humans, and to create new text using the same structure based on its "understanding". "Understanding" is only defined as the creation of a probabilistic model - it doesn't "understand" any meaning, only the structure of sentences and paragraphs, Yet, such a mindless machinary can demonstrate a wide range of interesting behaviors. Using the plane analogy: If you feed the airplanes designed by humans throughout history into GPT-3, GPT-3 is able to produce an original design of convincingly-looking airplanes of all nationalities and types. For example, if you say "1960, turboprop, Britain", it gives you one, and it's also capable of mixing things, such as putting a jet engine on a biplane, or designing a hypothetical WW2 Allies plane using Axis military technology. If you are lucky, some of the planes come with semi-functional systems. Some might be able to ignite its engine before it explodes, and few are even capable of taking off before it explodes (and some are unable to take off, but if you put them in the mid-air, it may be able to travel for one minute before it explodes). And it's all done under unsupervised learning, and without any domain-specific knowledge on Newtonian mechanics, aerodynamics or aviation. All it needs is an astronomical number of blueprints collection. To me it's an impressive achievement, even if it doesn't understand how to design airplanes.
  - m12k 6y ago
    I read somewhere that unlike GPT-2, GPT-3 "understands" addition. So if you feed it "74 + 129 = " it will come up with 203, even if that specific math problem never occurred in any of the text it was trained on. That to me is the most impressive result so far - not only is it possible, but also more efficient, for the model to actually encode the rules of addition instead of just memorizing the results for specific math problems. This to me seems like the dawn of insight and understanding, and I'm super excited to see what comes next.

lifeformed 6y ago

Well, what are you expecting? An actual explanation for the grand theory of everything? Or just a fun story as a response to the prompt? To that, it does a decent job, and that was all expected of it. Give a human the same prompt and see what they say.

leto_ii 6y ago

> It sounds nice but it's just one non sequitur after another. A very sleek looking plane that is incapable of flight. I like this analogy, especially because it alludes to the Dijkstra quote [1]. In this light it's probably unfruitful to try to create submarines that mimic swimming or that look like fish - you'll just end up with fancy looking stones. [1] "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim"

rrrrrrrrrrrryan 6y ago
Yes - in English, the verbs "swim" and "think" both imply that the action is being performed by a lifeform. "Fly," for some reason, does not. I wonder if this holds up in other languages as well.