5 ms·

LLMs don't make me question what we know about humans and thinking. They are really good at convincing us that they're good, but really, that's other humans bui

LLMs don't make me question what we know about humans and thinking. They are really good at convincing us that they're good, but really, that's other humans building stuff to convince us that it's good. There is no intelligence here, other than the perceived intelligence of predicting words intelligent people have written previously.

psytrancefan 2y ago
It does make me question humans and thinking but in the opposite direction. It is like sitting down at a piano, sight reading a piece from sheet music and then someone who has no idea what they are talking about claiming you composed the music on the fly. Then when you point out the sheet music they just double down on some bullshit as to why they are still right and that is still composing even though obviously it is not.
- ngneer 2y ago
  Best analogy so far. I am adopting this for the next wave of "wait until the next model" and "but humans hallucinate, too" comments. Yes, when we feed back our own output (language on the web) into ourselves, things become tricky to tease apart, and it would seem like intelligence to us. Then again, the mechanical turk appears intelligent, too. If we point out how it works, then the "magic" should vanish.
awongh 2y ago
> There is no intelligence here, other than the perceived intelligence of predicting words intelligent people have written previously. I think this is my main point- isn't it amazing that a thing that predicts words other humans have previously written manages to appear intelligent, or, more pointedly, have utility in communicating real thoughts and ideas? If you've ever asked an LLM a question and gotten a satisfying answer, that means that there is some human-level intelligence somewhere in the token filtering / recombinating that an LLM does. Specifically I think the test of human-like intelligence is literally the output- If we get utility from the arrangements of the tokens it outputs, that in and of itself demonstrates that some portion of human intelligence could be this same token generation mechanic.
- ngneer 2y ago
  No. Just means we are easy to fool. Like apes who see themselves in the mirror and fail to recognize they are seeing themselves in the mirror, thinking it is a different ape (and trying to mate with or attack it).
  - awongh 2y ago
    The invention of the mirror by humans probably provided an interesting insight into our own existence... I wonder what it would have been like to see your own reflection for the first time as a technology. How would that change your outlook on your self-hood and identity?
    - ngneer 2y ago
      Great question to ponder. Surely people would have seen their own reflections in water, but the mirror itself would have made "Reflection Technology" for "Artificial Introspection" more scalable. I suspect the mirror offered modern people a new viewpoint, allowing one to see how one is perceived by others. I do not think selfhood and identity would have been affected. My main question is about when people came to behave differently than apes. Douglas Hofstadter's "The Mind's I" may have a few hints on perception of self for you.
highfrequency 2y ago
> There is no intelligence here Can you list a few demonstrations from a text-outputting computer program that would each convince you that there is intelligence here? Eg writing a bestselling novel, proving an unsolved number theory conjecture, etc. Or is your belief uncontestable?
- esafak 2y ago
  I think he just needs the model to be installed in a humanoid robot.
- ngneer 2y ago
  That's not really a fair question. To answer it, the OP would have to define intelligence. If you have done so already, then by all means, do share your definition. If not, then you are in no better position to claim intelligence than the OP is in claiming lack thereof.

psytrancefan 2y ago

It does make me question humans and thinking but in the opposite direction. It is like sitting down at a piano, sight reading a piece from sheet music and then someone who has no idea what they are talking about claiming you composed the music on the fly. Then when you point out the sheet music they just double down on some bullshit as to why they are still right and that is still composing even though obviously it is not.

ngneer 2y ago
Best analogy so far. I am adopting this for the next wave of "wait until the next model" and "but humans hallucinate, too" comments. Yes, when we feed back our own output (language on the web) into ourselves, things become tricky to tease apart, and it would seem like intelligence to us. Then again, the mechanical turk appears intelligent, too. If we point out how it works, then the "magic" should vanish.

awongh 2y ago

> There is no intelligence here, other than the perceived intelligence of predicting words intelligent people have written previously. I think this is my main point- isn't it amazing that a thing that predicts words other humans have previously written manages to appear intelligent, or, more pointedly, have utility in communicating real thoughts and ideas? If you've ever asked an LLM a question and gotten a satisfying answer, that means that there is some human-level intelligence somewhere in the token filtering / recombinating that an LLM does. Specifically I think the test of human-like intelligence is literally the output- If we get utility from the arrangements of the tokens it outputs, that in and of itself demonstrates that some portion of human intelligence could be this same token generation mechanic.

ngneer 2y ago
No. Just means we are easy to fool. Like apes who see themselves in the mirror and fail to recognize they are seeing themselves in the mirror, thinking it is a different ape (and trying to mate with or attack it).
- awongh 2y ago
  The invention of the mirror by humans probably provided an interesting insight into our own existence... I wonder what it would have been like to see your own reflection for the first time as a technology. How would that change your outlook on your self-hood and identity?
  - ngneer 2y ago
    Great question to ponder. Surely people would have seen their own reflections in water, but the mirror itself would have made "Reflection Technology" for "Artificial Introspection" more scalable. I suspect the mirror offered modern people a new viewpoint, allowing one to see how one is perceived by others. I do not think selfhood and identity would have been affected. My main question is about when people came to behave differently than apes. Douglas Hofstadter's "The Mind's I" may have a few hints on perception of self for you.

highfrequency 2y ago

> There is no intelligence here Can you list a few demonstrations from a text-outputting computer program that would each convince you that there is intelligence here? Eg writing a bestselling novel, proving an unsolved number theory conjecture, etc. Or is your belief uncontestable?

esafak 2y ago
I think he just needs the model to be installed in a humanoid robot.
ngneer 2y ago
That's not really a fair question. To answer it, the OP would have to define intelligence. If you have done so already, then by all means, do share your definition. If not, then you are in no better position to claim intelligence than the OP is in claiming lack thereof.