4 ms·

Is it being lazy though? Or is it accidentally parroting an actual good mentor?

• orbital-decay 2y ago
  Not really, these assistants are all trained as yes-men, and the training usually works well. It might be a conflict between shorter outputs and the "soft reasoning" feature that version of Sonnet has, where it stops mid-reply and reflects on what it has written, in an attempt to reduce hallucinations. I don't know what exactly triggers it, but if it triggers in the middle of the long reply, it notices that it's already too long (which is an error according to its training) and stops immediately. (Or maybe I'm entirely wrong here!)