3 ms·

This is actually a very interesting idea, I hope someone would implement it and how. Would be fun to read the same text with stronger and stronger lossy compres

by GaggiX 3y ago

This is actually a very interesting idea, I hope someone would implement it and how. Would be fun to read the same text with stronger and stronger lossy compression.

dekhn 3y ago
up-goer five: explained using only the ten hundred words people use the most often
- ant6n 3y ago
  Carried to us by the Podcaster of Phd comics!
  - kbenson 3y ago
    What? Isn't it Randall Munroe (XKCD)?
    - dekhn 3y ago
      Oops- I left out the link! https://xkcd.com/1133/ https://xkcd.com/1133/ "you will not go to space today"
    - ant6n 3y ago
      Well, it was before that nnet compression.
- davidzweig 3y ago
  Oh, this would be super neat for (our) language learning chatbot.. restricting the vocab the model uses.. if someone knows how this can be done. :)
  - GaggiX 3y ago
    I guess you try to just zero the values that corresponds to unwanted tokens before the softmax.