Search: #TIL

Robotics is the least crowded high-value skill in tech right now The problem is that almost nobody knows where to start, because the field looks like five fields stacked on top of each other Some

AI Engineering Skills Map: Software engineering fundamentals

Someone recently asked me what advice I’d give …

The software development profession is in the midst of upheaval. Nobody knows how the AI revolution will play out in the end, but it is c...

Karpathy 讲如何成为专家、费曼讲如何检验真懂，本质是同一件事：用输出检验输入。这篇把「先做项目、再补理论、只跟过去的自己比」这套方法对应到神经网络的训练循环——Forward 是先动手，Loss 是和理想结果的差距，Backward 加梯度下降是复盘改进。人和网络最大的不同是面对大 Loss 会痛，所以关键是把「批判行为」和「否定价值」拆开：梯度下降只更新参数（行为、技能），别去炸掉架构（你是谁）。最后两个技巧：把错误序列化地写下来，然后立刻进入下一轮，别反复 Backward 同一个样本而过拟合。