阿斯塔纳时间
AI QAZAQSTANKZ · AI测试版

哈萨克斯坦人工智能与高科技领域的智能门户

Research1 分钟阅读

内容来自原始来源:

The Decoder

在 The Decoder 阅读全文

标题为英语OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles

OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose. It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks. Removing a flawed training dataset fixed the problem, but OpenAI…

The Decoder作者: Maximilian Schreiner2026年11月5日 (1小时前)
提供 1 种语言版本

AI 摘要:要点

  • OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose
  • It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks
  • Removing a flawed training dataset fixed the problem, but OpenAI…
Research
KZ·AI 8E60
Инсайт дня

OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles

标签:#OpenAI
Поделиться:

相关文章

Dropbox 以 3.5 亿美元收购 Tigris Data,以加强其人工智能基础设施

大型科技公司竞相锁定开发人员,Meta 收购人工智能编码初创公司 Conductor

Meta 正在收购人工智能编码初创公司 Conductor,该公司可让开发人员同时控制多个人工智能编码代理。同一天,OpenAI 宣布计划收购 Pierre Computer,Anthropic 已于 5 月份收购了 Steel。这些交易遵循一个模式:人工智能模型是……

The Decoder

返回列表