I wonder about that: if you're building public-facing chatbots then sure, you want everything as safe as possible – but from what I've seen the most valuable corporate use-case for LLMs is structured data extraction, where any denials are just a huge nuisance

→ View original post on X — @simonw,