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).