Comments by "Kristopher Driver" (@paxdriver) on "Stephen Wolfram: ChatGPT and the Nature of Truth, Reality u0026 Computation | Lex Fridman Podcast #376" video.

  1. Please do another episode with Wolfram very soon! Can't get enough.
    4
  2. 1:37:15 there is a deeper wisdom to the collective intelligence, as Lex says, but the internet is not a representative sample of the collective wisdom. The PhD and the idiots are both over represented by their publications and social media posts, but the common wisdom people share but don't publish are not proportionately incorporated into the training data. The risk of relying on gpt (not making freely available for general use, I'm saying rely on) is that there's no way for us to measure the quality of the human race's representation in the training set. The other issue that Wolfram misses on this same point is that we don't know how heavily the "average of the internet" is tokenized literally and how much is tokenized symbolically in its generative process, now or in future iterations. If tokens are deeply metaphysical and only superficially trained on tone, language) vocabulary and logic by the literal tokens then there's no reason gpt can't produce or prescribe new goals which we haven't considered consciously but which we're all working towards by cosmic evolutionary progression which we're still ignorant of but working toward (ie "God's plan", in another interpretation). Math has always been deeply linked to philosophy and art, I think we'd be wise to keep as wide a perspective of its potential influences because there are certainly facets of its application which we're overlooking in ignorance. That's why human play and prod, we discover deeper truths that way because our minds are wedged open to novel interpretations of the sandbox we play in.
    1