2026年03月27日 19:44:51
This was, Tom had come to understand, the core tension of the entire post-transition economy expressed in forty-five acres of vegetables. The AI systems were very good at general principles. They could optimize for a target, account for measurable variables, and respond to data faster than any human. What they couldn’t do was encode the kind of knowledge that accumulates over decades of physical presence in a specific place — the clay underneath the greenhouse, the deer path that compacted the soil in the northeast corner, the way the prevailing west wind dried the far rows faster than the ones sheltered by the tree line. This knowledge was in Carol’s head, not in any database, and it was precisely the kind of knowledge that natural-language specifications were worst at capturing, because it was embodied, contextual, and often inarticulable. Carol didn’t know that she under-watered the clay spot. She just did it. Her hands knew. The AI’s spec couldn’t capture what Carol’s hands knew, because Carol couldn’t put it into words, and words were the only thing the AI understood.
，更多细节参见搜狗输入法
Physical Intelligence的创始团队堪称机器人AI领域的“全明星阵容”，联合创始人均来自学术机构或科技巨头。
Bigoish: Evaluate the practical computational scaling of Rust algorithms
"We don't allow children to walk into bars or bottle shops, adult stores or casinos, but when it comes to online spaces... there are no such safeguards," Julie Inman Grant said.