Comment by orbital-decay

6 hours ago

I imagine Anthropic would rather train a small control model instead of resorting to sampling hacks

0 comments