3 ms·

If you're so sure about that then you should be able to provide a rigorous, testable definition of intelligence, and a proof of why that definition specifically

If you're so sure about that then you should be able to provide a rigorous, testable definition of intelligence, and a proof of why that definition specifically and nothing less is what must be met for machines to become exponentially self-improving.

dtagames 1y ago
It's the opposite. I can provide a rigorous description of how LLMs are built and operate (and have done so)[0]. There is no intelligence, however defined, in that process. It is the ultimate in rote behavior. Furthermore, LLMs are not learning or self-improving in any way. They are fed training data, corrections, and prompts and produce their answers solely through predictions, not as a result of learning. [0] https://levelup.gitconnected.com/something-from-nothing-d755f49d6636 https://levelup.gitconnected.com/something-from-nothing-d755...
- Asraelite 1y ago
  > There is no intelligence, however defined Ok, I choose to define intelligence as having the ability to transform vectors of tokens with matrices. There, LLMs are now intelligent and what you just said is incorrect, because this is one possible definition of intelligence. Seriously, nobody can argue against what you are saying if refuse to elaborate on what exactly you mean. Provide the definition for "intelligence" you are using.
  - dtagames 1y ago
    This recent post by another author is a good explanation that I agree with: https://news.ycombinator.com/item?id=44423206 https://news.ycombinator.com/item?id=44423206
    - Asraelite 1y ago
      That is some of the most nonsensical philosophical rambling I've ever seen. There are many problems with it but the main one is that it assumes a priori that humans are intelligent and machines are not, and then uses circular reasoning to justify that. > it is the human being reading intelligence into the behavior of LLMs. There is none in the LLM. There is no justification for why it must be a human and only a human that can "read intelligence into" something. I give up. I don't think you even know what you're claiming.

dtagames 1y ago

It's the opposite. I can provide a rigorous description of how LLMs are built and operate (and have done so)[0]. There is no intelligence, however defined, in that process. It is the ultimate in rote behavior. Furthermore, LLMs are not learning or self-improving in any way. They are fed training data, corrections, and prompts and produce their answers solely through predictions, not as a result of learning. [0] https://levelup.gitconnected.com/something-from-nothing-d755f49d6636 https://levelup.gitconnected.com/something-from-nothing-d755...

Asraelite 1y ago
> There is no intelligence, however defined Ok, I choose to define intelligence as having the ability to transform vectors of tokens with matrices. There, LLMs are now intelligent and what you just said is incorrect, because this is one possible definition of intelligence. Seriously, nobody can argue against what you are saying if refuse to elaborate on what exactly you mean. Provide the definition for "intelligence" you are using.
- dtagames 1y ago
  This recent post by another author is a good explanation that I agree with: https://news.ycombinator.com/item?id=44423206 https://news.ycombinator.com/item?id=44423206
  - Asraelite 1y ago
    That is some of the most nonsensical philosophical rambling I've ever seen. There are many problems with it but the main one is that it assumes a priori that humans are intelligent and machines are not, and then uses circular reasoning to justify that. > it is the human being reading intelligence into the behavior of LLMs. There is none in the LLM. There is no justification for why it must be a human and only a human that can "read intelligence into" something. I give up. I don't think you even know what you're claiming.