• Often times “perfect” is not achievable. Trying to win the perfect option results in losing the achievable “good enough” alternative. Learn to be satisfied with “good enough”.
• The complexity of a theoretical model can suddenly escalate by adding “just one more thing”, essentially making it intractable.
• Google Drive is not suitable for working with Office files. It works 99% the time, but that 1% failure results in loss of data, even in the cloud backup. Be uncool, use OneDrive instead.
• Even ideals should be realistic as life is all about compromises. What you’re looking for can’t check all the boxes in real life; if it does, it’s not attainable. In other words, if your goals only make you happy, you are missing something.
• Sometimes it’s better to choose a commercial software over an alternative FOSS (e.g., macOS vs. Linux). Companies stay around much longer than open-source communities. Open-source does NOT guarantee future-proofness.
• If you start with a cool tool and want to apply it to a problem in your field, you have already failed that research project. Start with topics, ideas, and problems.
  • Unless you have mastered the tool and your field.
• Large Language Models (LLMs) teach us something about our own inner-workings and thought-process. At the end of the day, we are essentially organic neural networks trained on data from our experiences and senses. Each person’s architecture is unique (due to, say, genetics), but no one is special. Learn to adapt to this reality.
• If you find yourself wondering what you should learn and how you must learn it, your mind is probably engaged in a self-defense mechanism. Thoughts like “should I learn Haskell?” and “which Haskell resource is the best?” are recipe for wasting time.
• When installing software or signing up on a website, first try to find your exit plan. Is the software easy to uninstall? Can you remove your account with one click? If the answer is “no”, don’t proceed.
• There is no point in accumulating “internet karma” on Hacker News, Reddit, etc. Be ok with getting downvoted but also be open to constructive criticism of your opinions by internet people. Karma just means how conforming you are, it doesn’t reflect your importance.
• If your opinions on one subject can be predicted from your opinions on another, you may be in the grip of an ideology. When you truly think for yourself, your conclusions will not be predictable.
• The best time to negotiate your salary for a new job is the moment AFTER they say they want you, and not before. Then it becomes a game of chicken for each side to name an amount first, but it is to your advantage to get them to give a number before you do.
• Python is fine. It’s just a nice glue language to make high-level abstractions on top of highly-optimized libraries that are written in C/C++. So you don’t lose much in terms of speed, but you gain much in terms of developer speed and ease of translating your thoughts to code. It took me 7 years to accept this reality and stop looking for alternative languages.
  • It makes more sense to become advanced in Python than to learn something else like Racket, Haskell, Clojure, Hylang, Julia, Common Lisp, Scheme, Nim, etc. and be mediocre at it.
• Labels could be dangerous. They [over] simplify things. Generalizing probably means fitting to noise.
• Always press “Update Score” when grading exams/quizzes on Canvas’ SpeedGrader, otherwise, all the scores will be lost…
• Egos and politics are as big of a thing in academia as in any other industry.
  • Yes, academia is an industry with its own perks and quirks.
    • No, I didn’t think this way until I embarked on my Ph.D. journey.
      • Yes, that takes away some of the joy of doing things just for the sake of science. “You gotta publish” means trendy ideas get you ahead while deeper thoughts might become your achilles’ heel because they’re not “sexy enough”.
        • No, that doesn’t mean someone who’s interested in certain areas should let them go for the sake of academic success.
          • Yes, it’s such a joy when you meet people who share this idea and are not merely robots who are good at ML stuff, but actually care about what they do.
• A paper with fancy machine learning methods (double ML, multimodal, etc.) is worth less than a comment on HN that actually teaches me something I didn’t know. Kudos to the authors of the paper for their endurance and patience to publish the paper though!
• Someone once told me they chose their job market paper topic based on what dataset was available to them. That was the saddest academic thing I’ve heard (and keep hearing). No passion for the topic, just “yet another project” they had to do to get a job. Sometimes I wonder if being passionate about something will ever be rewarded. It hasn’t been the case so far—obviously many people land prestigious positions through their fancy JMP. Let’s see what happens when I go on the job market!