The original link: Anthropic's April 2022 RLHF "helpful and harmless" paper

xennygrimmato_ · x · 2026-11-20

Companion post to the thread above: the author links Anthropic's April 2022 paper "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback". It was among the first to productize an RLHF-trained assistant optimized for helpfulness and harmlessness — an intellectual ancestor of the Claude line — though ChatGPT's launch later overshadowed it.

Related event: Revisiting Anthropic's 2022 RLHF Paper(2 posts)→

Original post →

More from Companies & People

Companies & People channel →