← Back to context

Comment by amoss

3 days ago

The tu quoque argument doesn't really work here. LLMs are specifically designed to output either material from their training set or instances that look statistically like it. Humans are not.

5 comments

amoss

Reply
lanstin  3 days ago

We aren’t specifically designed for anything, perhaps outside of the necessary consequences of having our mental capabilities and existing in an evolutionary context, so we can reason about tigers or integers and can persuade others to like us and have sex and help with our kids.

But our material natures are such that we have a profound need for external training sets - we don’t develop linguistic abilities unless we are exposed to a lot of speech. We can’t hear or produce phonemes in general that we are not exposed to during the training years.

Literacy, libraries and printing are so powerful because they extend the lifetime and size of our training material; if we each had to invent culture from our individual powers and skills, we would be living worse than anyone for fifty thousand years has lived.

encomiast  3 days ago

Sincere question: do we know enough about human intelligence to support that last assertion?

  • amoss  3 days ago

    I think it is uncontentious that we are not designed at all. But to address the intended spirit of the question: are we restricted to output that is a close fit for our training set?

    I would argue that we are not - because we do not have a rigid partition between training/test. Our ability to reason speculatively rests to some extent on ability to produce outputs that we have not seen before and which are statistically not like that which we have already seen. How we know which of these outputs to keep, and which to discard is (afaik) an open question.