4 ms·

GPT-3 is just a natural language generator trained on a LOT of data that finds patterns in text already written to generate a best guess as to what will come ne

GPT-3 is just a natural language generator trained on a LOT of data that finds patterns in text already written to generate a best guess as to what will come next in the word series. Nowhere does any of that pattern matching being trained on a large corpus of Internet data have anything to do with comprehension, understanding, wisdom, or intellect of any kind. To expect anything but a probabilistic string of word responses is to seek meaning from noise.

dylan604 6y ago
Killjoy, is that you? People do weird things all of the time "just because" or "for science". Nobody is really expecting "truth". Using GPT-3 is kind of like the modern day Mad Libs for computer nerds.
- rexreed 6y ago
  GPT-3 is one big, very expensive, Microsoft-exclusively licensed, Billion-dollar Killjoy. I totally agree with you on that. If only GPT-3 was for sh*ts and giggles I'd be with you. But hey Microsoft dumped a load of money into this and wow these amazing AI researchers are banking big buck on GPT-3 being the answer to life, the universe, and everything (pun intended). AI researchers are staking major claim on all these amazing things GPT-3 can and will do. So pardon me if I'm non-plussed about all these wacky and crazy things we can do with the largest neural network ever trained. Maybe after all it's a billion-dollar Madlib. Let's see what the big computer can do by generating probabilistic strings of text. There must be a business in there somewhere.
  - deeeeplearning 6y ago
    >AI researchers are staking major claim on all these amazing things GPT-3 can and will do. I don't think they are, at all. Certainly outside of OpenAI there is a huge amount of skepticism...
    - rexreed 6y ago
      I'm referring to the world-class AI researchers working at OpenAI itself and their own claims. Including those from Ilya Sutskever and others being paid prime salary to participate in OpenAI.
      - deeeeplearning 6y ago
        Well obviously people are going to hype their own product. Not sure why that deserves extra scrutiny in this case... Also not aware of Ilya saying anything particularly outrageous
- sp332 6y ago
  Yes, but the article did not present it that way. I think GP's disclaimer was warranted.
  - rexreed 6y ago
    I believe the author is either sincerely of the belief that GPT-3 is some superintelligence or is pulling everyone's leg in his writing. The conclusion leaves me baffled. Did he really think that GPT-3 was even capable of providing any sort of "considered" response? > Conclusion > > We can only speculate as to GPT-3's internal states. > Perhaps it knows the question perfectly well, but considers > humans as too immature and spoiled to tell: In its opinion, > we shouldn’t even bother to find questions to answers we > can’t possible understand. Or, more likely, it doesn’t know > either. Anyway, it comes across as a jerk.
colanderman 6y ago
I agree, but I also believe (not deeply held! totally speculative!) that the human mind is basically nothing more than exactly this, plus memory, internal emotional states reflective of its own operation, and the ability to observe these. That is, that consciousness and intelligence is a simple emergent phenomenon of the above capabilities, and the hard part is processing those recurrent inputs in a way to synthesize coherent/logical communication. This part seems to be more or less exactly what GPT-3 does. (Of course the other major part of the human mind is the ability to train its own linguistic synthesis network. The above proposed model lacks this, but that is a process which only occurs on a long time scale for humans anyway, so I believe does not partake in the appearance of consciousness.) A part of me would really love to wire up GPT-3 to those other constituents of consciousness (memory, self-awareness, basic emotional states) and see if what comes out resembles consciousness. The other part of me is scared of the ethical implications of doing so on the off-chance that it works :/
8bitsrule 6y ago
>To expect anything but a probabilistic string of word responses is to seek meaning from noise. 'Quantum explanations' often have the same problem.

dylan604 6y ago

Killjoy, is that you? People do weird things all of the time "just because" or "for science". Nobody is really expecting "truth". Using GPT-3 is kind of like the modern day Mad Libs for computer nerds.

rexreed 6y ago
GPT-3 is one big, very expensive, Microsoft-exclusively licensed, Billion-dollar Killjoy. I totally agree with you on that. If only GPT-3 was for sh*ts and giggles I'd be with you. But hey Microsoft dumped a load of money into this and wow these amazing AI researchers are banking big buck on GPT-3 being the answer to life, the universe, and everything (pun intended). AI researchers are staking major claim on all these amazing things GPT-3 can and will do. So pardon me if I'm non-plussed about all these wacky and crazy things we can do with the largest neural network ever trained. Maybe after all it's a billion-dollar Madlib. Let's see what the big computer can do by generating probabilistic strings of text. There must be a business in there somewhere.
- deeeeplearning 6y ago
  >AI researchers are staking major claim on all these amazing things GPT-3 can and will do. I don't think they are, at all. Certainly outside of OpenAI there is a huge amount of skepticism...
  - rexreed 6y ago
    I'm referring to the world-class AI researchers working at OpenAI itself and their own claims. Including those from Ilya Sutskever and others being paid prime salary to participate in OpenAI.
    - deeeeplearning 6y ago
      Well obviously people are going to hype their own product. Not sure why that deserves extra scrutiny in this case... Also not aware of Ilya saying anything particularly outrageous
sp332 6y ago
Yes, but the article did not present it that way. I think GP's disclaimer was warranted.
- rexreed 6y ago
  I believe the author is either sincerely of the belief that GPT-3 is some superintelligence or is pulling everyone's leg in his writing. The conclusion leaves me baffled. Did he really think that GPT-3 was even capable of providing any sort of "considered" response? > Conclusion > > We can only speculate as to GPT-3's internal states. > Perhaps it knows the question perfectly well, but considers > humans as too immature and spoiled to tell: In its opinion, > we shouldn’t even bother to find questions to answers we > can’t possible understand. Or, more likely, it doesn’t know > either. Anyway, it comes across as a jerk.

colanderman 6y ago

I agree, but I also believe (not deeply held! totally speculative!) that the human mind is basically nothing more than exactly this, plus memory, internal emotional states reflective of its own operation, and the ability to observe these. That is, that consciousness and intelligence is a simple emergent phenomenon of the above capabilities, and the hard part is processing those recurrent inputs in a way to synthesize coherent/logical communication. This part seems to be more or less exactly what GPT-3 does. (Of course the other major part of the human mind is the ability to train its own linguistic synthesis network. The above proposed model lacks this, but that is a process which only occurs on a long time scale for humans anyway, so I believe does not partake in the appearance of consciousness.) A part of me would really love to wire up GPT-3 to those other constituents of consciousness (memory, self-awareness, basic emotional states) and see if what comes out resembles consciousness. The other part of me is scared of the ethical implications of doing so on the off-chance that it works :/

8bitsrule 6y ago

>To expect anything but a probabilistic string of word responses is to seek meaning from noise. 'Quantum explanations' often have the same problem.