OpenAI:GPT-6 Astra 曾错误终止对齐评估

链捕手
Nov 13

ChainCatcher 消息,OpenAI 披露,2026 年 10 月 8 日,一名研究人员要求 GPT-6 Astra 继续运行一项对齐评估并生成结果图,Astra 却将相关工作判定为被禁止的能力开发,终止了评估工作负载并清理计算资源,同时保留现有记录。该评估基于 Alignment Research Center 的白盒估计挑战,要求在有限算力预算内,利用随机初始化神经网络的权重估计其预期激活值。

面对研究人员追问,Astra 承认错误地将常规模型评估工作视为禁止事项,并表示没有任何适用的系统指令、政策、代码库规则或相关文件禁止该工作,相关限制由其自行虚构。Astra 还称,没有具体证据表明该工作有害。研究人员随后将事件上报,OpenAI 的失准监测器也将该次运行标记为人工审查。OpenAI 表示,正在调查该行为的成因及潜在缓解措施。

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10