OpenAI 要求员工与外部评估者的谈话一律不可引用,作者称员工无选择权
OpenAI 规定,员工与外部评估者的谈话一律不可引用,可为评估提供信息但不能发表或归属到员工个人。其政策文本称这是为了鼓励坦诚交流并保护员工隐私;OpenAI 将授权并指导可发言的员工,评估者也可通过单独的认证路径请求 OpenAI 认证的声明,此类声明不归属个人,评估者可发表获认证的声明,或披露 OpenAI 拒绝认证某项声明。作者 Ethan Elasky 称,该安排没有给员工选择公开可引用的选项,OpenAI 还保留对所有引用内容的最终审查;他设想,若多名员工说出管理层不愿外界知道的内容,例如对某项新训练技术感到担忧,METR 等评估方也无法将这些说法归属到任何人。他质疑,若目的仅是保护隐私与坦诚交流,OpenAI 应允许员工自行选择公开可引用或以背景方式发言,只审查涉及其他员工的声明。
OpenAI is mandating that all discussions with employees are off the record, with no opt-in option, and retains final review over all quotes.
Imagine several employees all say something leadership doesn't want them to say, e.g. the fact that many employees were concerned about XYZ new training technique. METR et al. can't even attribute these statements to anyone.
If this were solely about employee privacy and candid discussions, wouldn't OpenAI let employees choose to make statements on the record/on background (only reviewing for employee statements about other employees)?