Most Important Traits for AI Research

Feedback is like a gradient, it tells you which direction to go to become a better researcher.

Jason Wei is an AI researcher working at OpenAI who popularized chain-of-thought prompting.

He shares some advice for research, specifically what is the most important trait for research.

He views research as a skill that can be practiced, like sports or music.

The short answer is grit, since grit enables practice.

An underrated skill is learning from feedback.

Take very seriously the feedback that others give, especially if it's coming from someone at the top of their game.

Ask your mentor/advisor every week what you can do better, and do it.

It's also good to know when to ignore feedback. They might not have thought about your problem as much as you have.

Another skill is to be willing to do gruntwork, especially looking at data. Jason had to spend 40+ hours to do data labelling and validation.

The reality of research is a lot of hard work and drudgery, and it's an important part of the research process. It can help you build intuition about the research.

The final skill is being a good communicator.

Communicating well makes you more trustworthy to work with.

Don't make promises you can't keep, be realistic about when you can meet deadlines and finishing tasks.

Be specific when communicating blockers, instead of "X doesn't work", say "X didn't work when I tried it using formulation F on model M using dataset D" which decreases the mental load of the other person.