Y
Hacker News
new
|
ask
|
show
|
jobs
by
disgruntledphd2
112 days ago
It depends on how much context is in the training data. I find that they make stuff up more in places where there isn't enough context (so more often in internal $work stuff).