First year

General tips:

• Write a bullet point outline of the paper early on and fill in formal artifacts and key plots quickly.
• Deadlines can create useful pressure for me.
• I should maximize information gain / importance / chance to kill the project with experiments and the research questions I ask.
• Thoroughly read and understand related literature.
• Write like Lawrence Saul (use the folder and CLAUDE.md).

Producing research that Matters:

1. Talk to the consumers of your research before the results exist.
2. Write the theory of change early on.
3. Don’t make projects bigger than they need to be and do a few things well
4. Maybe try shipping intermediate artifacts publicly
5. Last mile delivery

Work on important problems, where “important” means consequential and you have a reasonable attack. Keep a running list of 10–20 such problems and pounce when a new idea gives you a way in. Regularly set aside time to ask what the important problems are, and if your beliefs and your daily work point in different directions, change one of them. Stay exposed to the best people you can find, mostly through one-on-one conversations about half-formed ideas with people who push back rather than just agree, and use those conversations and your reading to learn what the problems are, not to borrow solutions. Commit deeply, spend serious time selling the work, and don’t make excuses to yourself.

I wanna get better at thinking and doing research and here’s how I’m going to try to do so.

• Pay attention to the strength of arguments and the argument structure. Are they soely deductive claims? Do they withstand argumentation? Are assumptions justified? Reminder, Deduction, induction, abduction. Deduction is just pushing facts, induction is generalizing, and abduction is expanding the hypothesis space. Claude gave a good example:
  • Deduction. Rule: all beans from this bag are white. Case: these beans are from this bag. ∴ Result: these beans are white.
  • Induction. Case: these beans are from this bag. Result: these beans are white. ∴ Rule: all beans from this bag are white.
  • Abduction. Rule: all beans from this bag are white. Result: these beans are white. ∴ Case: these beans are from this bag.
• Consider framing research as an RL problem with sparse reward on some data.Obviously I want to create a rich feedback signal on high quality data. In research this means getting lots of feedback from myself and others, and consuming content that I can build good abstractions from. More specifically I can
  • Look at a paper problem statement and abstract and predict what their central thread and main results are
  • Cleanly break papers down into individual argument steps and reconstruct the paper from just those steps
  • Ask for feedback on my ideas and experiments often and before everything is finished.
  • Make connections between papers
• Goal directed research a la John Schulman
• Take ownership and turn they into we. I thought they were thinking about xyz problem… No, it’s we!

Researchers who have developed good taste find clever and elegant approaches, and find themselves pulled toward solutions that feel “right” before they can fully articulate why. At a macro scale, they pick problems that will matter; at a micro scale, they take approaches likely to succeed. And perhaps most importantly, they make these decisions early, before wasting months of effort.

But if I had to summarize it in one sentence, it would be that taste comes from practicing the skill of research, keeping your focus always on identifying what works and what doesn’t.

Once you’ve read everything, the second step is to forget it all.

Hamming was famous for asking “what’s the most important problem in your field, and why aren’t you working on it?” Approach your research with this mindset. One excellent paper is worth a thousand mediocre ones, and takes less time to write.

So the magnitude of your contribution is judged, in a very real sense, by counting the months between when you publish, and when the next person would have. Try to pick something that would have taken at least a few months for someone else to do as well as you did

Knowing what I’m looking for by going off rigorous falsifiable hypotheses as I’m doing my search as much as I can

https://nicholas.carlini.com/writing/2026/how-to-win-a-best-paper-award.html

https://colah.github.io/notes/taste/

Make it concrete! When pitching an idea what’s a specific experiment I could do / theorem I could prove to verify it?