How to Become World-Class AI Researcher?

Shelvia
3 min readApr 19, 2024

--

I watched an interview with Sholto Douglas (AI researcher at Google DeepMind) and Trenton Bricken (AI researcher at Anthropic), and have been greatly inspired by what they said. Here are some reflections:

Many Ideas, Need Execution

Trenton mentioned that he and his team constantly have many ideas, but all these ideas need careful, thorough, and quick investigation. I totally relate to this point. Sometimes when my labmates and I discuss some problems, we come up with a bunch of ideas on what kind of interesting analysis we can do. Sadly, many of these ideas stay as ideas since neither of us has the time nor energy to execute them. There’s no excuse for this, of course. We all need to get better at quickly executing any good ideas, which is a very important trait to have (Sholto called it the agency at work). Good ideas are nothing without proper execution. The implementation also needs to be fast (without sacrificing correctness) because many good ideas are lining up, waiting to be tested out there.

Pursue a Problem till the End

Both Trenton and Sholto agreed that it’s very important for a good researcher to have the dedication to pursue a problem until the end. If you encounter a problem, don’t simply give up or get stuck on it for a long period of time. Instead, really delve deep into the problem itself and conduct your own research on how to potentially solve it. Maybe talk to people or different experts to get some ideas on how to overcome each problem. No matter what, there’s no excuse for just saying you are not good enough or the problem is simply too hard. Even if it’s a technical issue that you don’t have expertise in, you can still learn about it and potentially acquire new (and useful) skills after solving it. So don’t give up!

Getting Good at Picking Problems

Sholto mentioned that one of the ways he has been impactful in his team is by being very good at picking problems that haven’t been well-solved. If you pick a problem that is very “hot” in the field right now, you are competing with the very best, and they potentially have more resources to solve the problem. On the other hand, if you pick a relatively unexplored but important problem to solve and manage to solve it, that alone can make you famous. This again requires tenacity and dedication to stick to solving the problem, no matter how hard it gets. The ability to pick a hard but solvable problem requires one to read widely and build intuition about what works and what doesn’t.

Manufacture Your Own Luck

Being recognized as world-class AI researchers is all about being in the right place at the right time. You could call it luck. But truthfully, you can manufacture your own luck (or manifest it better) by putting yourself out there. Engaging in independent learning, attending conferences, and networking with various researchers greatly improve your chances of being recognized. When you do some independent learning outside your scope of work, you are training yourself to become a general problem solver. You could potentially offer a fresh perspective on looking at a problem based on your experience from a different field. This also allows you to talk to a wide variety of researchers and opens up more opportunities for you.

Caring Unbelievable Amount

Sholto mentioned at the end of the video that above everything else, the most important attitude to have is caring an unbelievable amount about a problem. When you do, you don’t give up on the problem easily, you try to think of any possible solutions to overcome any obstacles that you encounter, and you make sure the investigations you carry out are thorough and correct. You naturally want to improve yourself so that you can be better at solving these problems, and you also become more detail-oriented, carefully checking that every step of the implementation is correct. I guess this is equivalent to being passionate about what you are doing, and many good AI researchers undoubtedly have this quality.

Image by the author using DALL-E 3.

Thank you for reading! I hope this article has inspired you as much as it did for me. Let me know in the comments if you have any thoughts to share! Let’s all strive and work hard to become better researchers! :)

--

--

Shelvia

Researcher in Information Theory and Trustworthy AI. Addicted to puzzles and brain teasers. Interested in particle physics and neuroscience.