Spaces:
Running
Running
The model makes reproducible mistakes in understanding logical symbols in the input.
#16
by
alxfed
- opened
It calls the negation symbol ¬ a 'caret'. It says that turnstile symbol ⊢ is 'negation' and on an on like that (sorry, don't have time to test all of the symbols). Basically the encoding in the training dataset for logical symbols was wrong and now the model misunderstands what a user is typing.
The right symbols and their encoding are here: https://en.wikipedia.org/wiki/List_of_logic_symbols