@Xudong07452910: I recently came across a long post by Anthropic researcher Vivek (@itsreallyvivek) on how to truly train your research ability. What struck me most: many times we think we're doing research, but we're really just chasing trends, scrolling Arxiv, seeing what others are discussing, and then picking up those problems.

Recently, I came across a long post by Anthropic researcher Vivek @itsreallyvivek about how to truly train your research skills.

What struck me the most is this: often, we think we’re doing research, but actually we’re just chasing trends, scrolling through Arxiv, and following what others are discussing, then picking up those problems and continuing.

This can certainly produce results, but it easily turns into a kind of role-playing of science: looking very busy, consuming a lot, taking many notes, running many experiments — yet deep down, you’re not really sure what you’re trying to ask.

Vivek’s post reminded me that research ability isn’t some mystical talent — it’s a set of very down-to-earth habits: choose your own problems, predict the outcome before running an experiment; read the original papers; write down your hypotheses, results, and changes in thinking; stare at failure cases, not just the loss curve.

The longer I go on, the more I feel that doing research is really like training yourself.

Whatever input you feed your brain, it will develop the corresponding taste; whether you’re willing to face mistakes determines how fast you improve; whether you write things down decides whether you’re honestly iterating or just relying on memory to whitewash yourself.

Homogeneous input only leads to homogeneous output.

What truly makes the difference may be who can, over a long period of time, slowly train their own sense of problem, judgment, and error-correction system.

This post is worth reading again when you feel lost.