10 Matching Annotations
  1. Apr 2026
    1. 未来的 CNC 也许不是一团越来越大的连续表征,而会更像一套可路由、可组合、局部更容易检查的机器底座。

      这一观点挑战了当前AI模型向更大规模发展的主流趋势。作者提出神经计算机可能更接近离散、稀疏、局部可验证的结构,这暗示了AI发展可能存在与当前大模型路线完全不同的方向,具有颠覆性意义。

  2. Feb 2024
  3. Jul 2021
    1. Now let’s consider an alternative situation: the BSCU Autobrake provides two separate discrete control actions to Start Braking andto Stop Braking

      A priori this seems superior (or just simpler?) to me. Perhaps worth asking a clarifying question in class: why would you prefer continuous vs discrete actions in different scenarios?

  4. Mar 2021
  5. Jan 2021
  6. Jul 2020
  7. Jun 2020
  8. Sep 2016