Hacker News new | ask | show | jobs
by orbital-decay 4 days ago
I imagine Anthropic would rather train a small control model instead of resorting to sampling hacks