Hamming's Research Insights | Generated by AI

Home 2026.07

Question: What does Richard Hamming’s talk “You and Your Research” say?

Answer:

Hamming gave this talk at Bellcore in 1986, drawing on his time at Bell Labs alongside Shannon, and the transcript has become required reading in serious research and startup circles precisely because it’s not motivational fluff — it’s a mechanistic breakdown of why some people produce great work and most don’t, even when talent is roughly equal. Given your trajectory (leaving a stable bank job to go all-in on frontier AI work), this talk maps unusually well onto your situation. Key points:

1. Luck is not the differentiator — but “courage” and preparation are

Hamming opens by dismantling the “great scientists just got lucky” narrative. His claim: luck favors the prepared mind, and preparation means you’ve already done the deep thinking on important problems before the opportunity shows up. This is directly your “learning while earning” thesis — training GPT-2 from scratch, working through nanochat, going deep on DeepSeek v4 MoE before clients ask for it. When a high-value client appears, you want to already have the depth, not be learning it on their dime.

2. Work on important problems

His most quoted line: most people work on problems they know how to solve, not on problems that matter. He’d ask colleagues “what are the important problems in your field?” and then “why aren’t you working on them?” Usual answers: no good approach exists yet, or it seems too hard. Hamming’s point: important-but-hard problems are exactly where the compounding returns live — the same effort spent on a trivial problem yields little, the same effort on an important open problem (even if you fail) yields skill and reputation others can’t buy.

For you: this argues for staying on MoE training internals, RLHF/GRPO, and inference optimization rather than only shipping “integrate GPT-4 API into a CRM” consulting gigs. The API-wrapper work pays bills; it’s not what compounds into a moat. You’ve already identified this correctly in your own playbook (#2 — “Tinker-level skills as moat”).

3. Open doors vs. closed doors

Hamming distinguishes people who leave their office door open (interruptible, but exposed to serendipitous conversations, collaboration, and outside signal) from those who close it (focused, but isolated from what’s actually happening in the field). He argues the open-door people, over a long career, out-produce the closed-door people because research is not a solitary activity — the important problems and their solutions surface through contact with other people’s problems.

Modern translation for you: build in public (which you’re already doing via your notes and blog) is the open door. It’s not just marketing — it’s how you stay exposed to what problems other builders are actually hitting, which is where your next consulting lead or product idea comes from.

4. “The Drunkard’s Walk” — small, favorable-direction increments compound

Hamming’s model of scientific progress: not big undirected leaps, but a walk with a slight bias in a useful direction. Over enough steps, that small bias dominates. This is literally the compounding logic in your own long-term plan — every client project funds deeper AI research, every model trained is one more push toward Agentic-world infrastructure. Hamming would say: make sure the bias is real, not imagined. Ask honestly whether today’s work moved you toward the important problem or just kept you busy.

5. Selling your work matters as much as doing it

He’s blunt about this: brilliant work that isn’t communicated, packaged, and delivered in a form others can use might as well not exist. Style, communication, and salesmanship are not “beneath” real researchers — they’re part of the job. This maps to your “Launch channel: AI · Live” strategy — the quality of what you write and ship publicly is the moat-building mechanism, not separate from it.

6. Emotional commitment and the 10%-more effort

Hamming observed that the top researchers at Bell Labs weren’t necessarily smarter — they worked with a different intensity and had an emotional commitment to solving specific important problems, often over years. Comfort and average effort are the enemy of great work, not lack of raw ability.

Where the talk is honest about the cost

He doesn’t dodge the tradeoffs: he explicitly says great researchers often sacrifice family time and are hard to live with, and he’s clear-eyed that this is a genuine cost, not something to romanticize away. Given your situation — wife and parents not on board with the startup leap — this is the part of the talk worth sitting with rather than skipping. Hamming isn’t selling “no pain, all gain.” He’s saying the tradeoff is real and the top people made it consciously, eyes open, and it’s worth being honest with yourself (and them) about the price rather than pretending there isn’t one.

Most quotable framing for your context: Hamming’s core diagnostic question — “What are the important problems in my field, and why am I not working on them?” — is worth running against your own roadmap monthly. Right now your honest answer seems to be “I am” (MoE, GRPO, agent architectures), which is the rare, correct answer. The risk for you isn’t lack of ambition — it’s letting service-work gravity pull you back toward safe, billable-but-shallow API integration instead of the frontier training work that’s your actual moat.

References:

Back Donate