Further, OpenAI will now notify authorities if it detects “imminent and credible” threats in ChatGPT conversations, even if the user doesn’t reveal “a target, means, and timing of planned violence.” O’Leary explained that if the new rules had been in effect when the shooter’s account was banned in 2025, the company would have notified the police. OpenAI will also establish a point of contact for Canadian law enforcement so it can quickly share information with authorities when needed.
宋健：这个Benchmark理论上是随基模能力进化、垂直场景的强化学习，针对性的工程Infra打磨来定的，小龙虾还是有下降空间。再加上基模本身能力上升100倍，成本100倍下降也是有可能的。
。新收录的资料对此有专业解读
Soucolor 9x12 Sketchbook
▲ Lightwear AI 耳机
Архивное фото. Фото: Steve Lynes / Wikimedia