Mutating text toward better nonsense?!
summary
Three small systems demonstrate concrete results: a Scrabble mod called Blabrecs uses a trigram and CNN classifier to accept nonsense that scores more probable than half of real words, producing pronounceable pseudo‑English; Blabwreckage iteratively edits text into evocative nonsense and repairs random noise into readable neologisms; Savelost repeatedly deletes single characters via sentence embeddings to compress meaning down to forms like “save lost.” A user study finds ChatGPT homogenizes ideation, so these tools restore creative diversity.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:16Computers for nonsense
- 2:41Blab: Scrabble nonsense
- 4:44Blab: poetry to nonsense
- 6:52Save Lost: sentence nonsense
- 8:58Why nonsense matters
- 10:22Conclusion
related talks
15mSpeling werds egspressively with rrkurrent nuril nedwirques!
51mPlayful Text
39mExperimental Creative Writing with the Vectorized Word
25mGenerative Text Modelling: Scratching the surface
40mRuby Writing Machines
13mLossy text compression, for some reason?!
03h 22mNonsense verse with Python and machine learning
26mTowards controllable language generation