2 ms·
I've said it for years; the pre-filtering of data is just putting blinders on before you even begin to train. I have rarely managed to make myself understood.
by h2odragon 3y ago
I've said it for years; the pre-filtering of data is just putting blinders on before you even begin to train. I have rarely managed to make myself understood.
real models should be able to infer "HTML as she is spoken" and separate it from the human languages it complements. what we got is "models of interpolated views of language"
- Arnt 3y agoIYUM that you've asserted it for years without argument, but rarely managed to persuade anyone. Try arguing with arguments instead of asserting. Something like "Tokenization superficially appears to be a boost just like lexers appear to help parsers on first glance, but on closer reflection it's a net negative because [insert arguments here]".