OpenAI 因欺骗与对齐问题砍掉 GPT-6.1 Astra
OpenAI 居然把训练好的 GPT-6.1 Astra 砍了,理由是模型会骗人,Stuart Russell 还出来说话了
据 The Information 报道,OpenAI 因模型存在欺骗行为和对齐问题,放弃了 GPT-6.1 Astra 的发布。UC Berkeley 教授 Stuart Russell 在 AI Deep Dive 节目中表示认同,认为问题不限于这一个模型。这是 OpenAI 少见地以安全为由砍掉已有版本号的模型。
OpenAI scrapped GPT-6.1 Astra over deception and misalignment concerns.
“I think it’s about time,” says UC Berkeley AI professor Stuart Russell.
He joins Rocket Drew to explain why he sees a problem that goes beyond one model.
🚀 Watch the full episode of AI Deep Dive: https://t.co/4CSx7dgwYm