Loading…
9
Just realized people keep feeding AI models clean data and forgetting commas in the training set
Spent 3 weeks testing a small language model in my spare room and found that stripping punctuation made it way worse at reading dates and lists. One quick fix with 500 extra labeled sentences made its answers way more on point. Am I the only one who missed how much punctuation matters for training?
1 comments
Log in to join the discussion
Log In1 Comment
the_harper19d ago
Guessing you learned that one the hard way after three weeks of talking to a robot that thinks March 5 2024 is a phone number. Punctuation is basically the difference between "let's eat grandma" and "let's eat, grandma" and your model was out here eating grandmas.
4