The job of AI researchers is to teach machines how to "think"
Bitter lesson: progress of AI in the past 70 years boils down to :
- Develop progressively more general methods with weaker modeling assumptions
- Add more data and computation (i.e. scale up)
The more structure, the less scalable the method is。但是如下图，可能在没见到效果之前，就先放弃了
