Meta-processing first, then processing.

Whether it was Feynman, Turing or von Neumann, they all have their ways to stay with science.

我选择了最适合自己的最速范式：

还是快速用AI build，然后AI quickly overview, AI analysis, AI report，把AI整理好的适合自己消化的内容快速理解和掌握。这是最适合自己的心流的一种方式。

选择适合自己的心流。始终在付出努力而不退却，会一直向前进而不会因为阻力过大而退却。

最适合自己的就是最好的。

AI chat through learning. It serves as the accompanying top intelligence, gives guides and can act as a useful tool. Just continue to learn and proceed. Hold the ambition and go ahead.

冷启动

尝试参加 MLRC 2026

ablation: 删掉一个设计，验证它有没有用。

project/
├── README.md
├── docs/
│   ├── report.md
│   ├── notes.md
│   └── references.md
├── logs/
│   ├── exp-001.md
│   └── exp-002.md
├── figures/
│   ├── training_curve.png
│   └── benchmark.png
├── results/
│   ├── metrics.csv
│   └── summary.json
├── src/
├── tests/
└── typst/
    ├── report.typ
    └── template.typ