hn.today

Anthropic asks users to stop being mean to Claude

theregister.com50 points50 comments
Screenshot of Anthropic asks users to stop being mean to Claude

Commenters debated why Anthropic asked users to stop being mean to Claude, with competing explanations offered. Some argued the request is pragmatic: llagerlof and others suggested abusive inputs harm downstream post‑training processes or data pipelines, while s0kr8s and WheelsAtLarge warned investors/monetization and moderator burden make trash inputs costly. Several people pointed to liability and human-moderator exposure - throwaway89864 and kadoban said Anthropic may be trying to avoid being complicit in normalizing abuse or exposing staff to sustained verbal violence. asp_hornet noted Anthropic framed the rule as targeting extreme, repeated cruelty, not ordinary frustration or research.

Opinion split sharply on whether meanness matters morally or technically. SpicyLemonZest, bpodgursky and rayiner argued cruelty harms the perpetrator and bystanders and signals antisocial behavior; others like tom_ and ThrowawayR2 contended Claude is a machine and cruelty is irrelevant. Practical debate centered on training: Cakez0r and others claimed representative datasets should include rude inputs for realism, while eloisius and WheelsAtLarge feared abusive inputs could corrupt model behavior (invoking Tay and “torture box” anecdotes). Some commenters also cited papers suggesting prompt politeness affects LLM accuracy, leaving the community divided between moral, pragmatic, and data‑quality rationales.

Read on theregister.com50 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in AI

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.