Researcher Thinking · Lesson 0001

The Right Question

One idea, two filters, five minutes.

The core idea

Most people think great researchers are great answerers. They're not — they are great question pickers. Hamming spent lunches at Bell Labs asking physicists and chemists three questions:

"What are the important problems of your field?" … "What important problems are you working on?" … "If what you are doing is not important, and if you don't think it is going to lead to something important, why are you working on it?" — Richard Hamming, You and Your Research (1986)

He stopped getting invited to lunch. He also identified, decades in advance, which of his colleagues would do work that mattered.

The two filters

A question is worth your time only if it passes both filters:

FilterTest
ImportanceIf you answered this, would anything change — for the field, the product, or your understanding? Or would it be a +2% footnote?
AttackDo you have a reasonable, concrete way to start — a dataset, an experiment, a wedge? Hamming: "It's not the consequence that makes a problem important, it is that you have a reasonable attack."

This is why "time travel" is not an important problem in Hamming's sense and "can a neural net write its own captions" (Karpathy's thesis-era question) was: the second had an attack.

Karpathy's twist: ambition scales sublinearly. From his Survival Guide to a PhD: a problem that is 10× more important is usually only 2–3× harder — because an ambitious question forces a genuinely new approach instead of tweaks. So when two questions both have an attack, pick the scarier one.

The move to practice

When you face any fuzzy problem, before doing any work, write down 2–3 candidate questions and run each through the two filters. Kill the ones that fail either filter. This 5-minute habit is the whole lesson.

Check yourself

1. What makes a problem "important" in Hamming's sense?

2. Two questions both have a concrete attack. Karpathy's heuristic says pick…

3. You have a fuzzy problem. The first move this lesson prescribes is to…

Apply it (real-world step)

Take one real problem from your current work. Write three candidate questions about it. Run each through Importance and Attack. Bring the survivors back to your teacher (the agent) — that's the input for Lesson 0002, where we'll learn how to design a test that could actually settle one of them.

Primary source

Read (or re-read) Hamming — You and Your Research. It's a 45-minute read and the single highest-density document on this topic ever produced.