The Goblin in the Machine: How OpenAI’s Weirdest Bug Became an Alignment Warning
中文摘要
OpenAI发现Codex生成文本时会不自觉地提及奇幻生物,需限制相关内容。
English Summary
OpenAI had to restrict Codex from discussing various creatures, a strange bug that serves as a cautionary warning about the complexities of AI alignment.
Original Excerpt
OpenAI had to tell Codex not to talk about goblins, gremlins, raccoons, trolls, ogres, or pigeons. Continue reading on Write A Catalyst »