3 ms·
I generally seed with one word, but I could in theory use as many as there are memory-steps in the LSTM. For the shorter poems, I run with 50 steps worth of mem
by karmel 10y ago
I generally seed with one word, but I could in theory use as many as there are memory-steps in the LSTM. For the shorter poems, I run with 50 steps worth of memory, so could reasonably seed with up to 50 words, and the machine would carry on from there.
As to multiple meanings, I do not force an interpretation, and the model is simple enough that it does not explicitly disambiguate. Part of the fun for me as an artist is letting the model go where it may, often surprising me when it does.
- zebraflask 10y agoWith up to 50 words, it sounds like it would be a great fit for haiku and similar short form pieces.
- karmel 10y agoAgreed, but a haiku requires some notion of syllables. Other machine poets on CuratedAI do work with more complicated syntactical structures (I.e. http://curatedai.com/poetry/three-haikus-in-honor-of-haruki-murakami/ http://curatedai.com/poetry/three-haikus-in-honor-of-haruki-... , http://curatedai.com/poetry/mildly-concerned-the-woman-next-to-me-just-prayed-for-me/ http://curatedai.com/poetry/mildly-concerned-the-woman-next-... ), though my personal bias is more free form, a la e e cummings.
- zebraflask 10y agoEven free form has to have a little structure, don't you think? Know rules before you break them? How do you think it would handle Milton?