"PrismML 称 1-bit Bonsai 27B 只需 3.9GB、在 iPhone 上达到每秒 11 token，并保留 FP16 平均分的 89.5%。这几个数字分别意味着什么？"
"PrismML says its 1-bit Bonsai 27B occupies 3.9GB, runs at 11 tokens per second on an iPhone, and retains 89.5% of FP16 performance. Here is what each
OpenAI Presence 把策略、批准动作、仿真、评测、人工升级和受控更新连成生产闭环。本文拆解平台应该负责什么，以及企业不能外包哪些控制权。
OpenAI Presence turns policies, approved actions, simulations, evaluations, escalation, and controlled updates into one operating loop. Here is what t
"五项实验显示，错误 AI 建议会降低准确率并抬高信心。真正可执行的应对，是把独立判断和外部核验做进工作流。"
"A five-experiment preprint shows how wrong AI advice can reduce accuracy and inflate confidence. The practical answer is a three-stage verification l
Anthropic 在 Claude 内部发现了一个低容量、可语言化且具有因果作用的 J-space。它能承载没有输出的中间概念，却不能证明 Claude 拥有主观意识。
Anthropic found a small, causally active J-space inside Claude. It carries silent concepts but does not prove subjective consciousness.
"OpenAI B2B Signals 报告显示前沿企业每个员工的 AI 智能消耗是典型企业的 3.5 倍。36/64 的体积与深度分裂揭示了为什么大多数企业在 AI 采用上问错了问题。"
"OpenAI's B2B Signals report reveals frontier firms use 3.5x more AI per worker than typical firms. The 36/64 split between volume and depth reveals w
