I’ve just started a job as a data scientist at Recursion Pharmaceuticals. I’m using machine learning to find new drug compounds.

The basic model is:

  • take some rare diseases that are caused by single genes;
  • simulate these diseases cheaply and at scale with siRNA knockdowns;
  • detect (here’s the machine learning part) how images of sick cells look different from healthy cells
  • observe (machine learning again) which drugs make sick cells look like healthy cells
  • send the promising drugs on to in-vitro and in-vivo screens

This is basically my dream job. I’ve been torn between math and biology since I was maybe 9; now I get to do both. And I get to work towards precisely the problems I care about: curing diseases, getting as much purchase as possible out of computational methods in practical applications, reversing Eroom’s Law, etc. I’m thrilled to be working at Recursion.

Due to company policy, I won’t be able to continue doing freelance lit review any more, at least not for paid projects; I expect to keep doing the occasional free project here and there.

I’ve also made a few updates in my views recently that I thought I’d share here.

  • The boost in my productivity and overall well-being from having a meaningful job, working with people I trust and respect on problems I care about, is enormous. Much more than I’d have expected. I am now much more sympathetic to messages like “too many people are trapped in bullshit jobs”, “pointless busywork in school is harmful”, “it’s bad to be alienated from one’s labor”, etc. I’m more bullish on things like self-employment, unschooling, quitting your job to pursue your passion, and so on; stagnation is a real cost to your soul.
    • I’m reminded of the theories of people like Gabriel Kolko, who said that the bigness of “big business” is an artifact of regulatory capture, in which large businesses are subsidized by the state. In this model, the “natural”, undistorted size of businesses would be smaller, and fewer things would be done that had no real purpose besides checking an officially-required box. Pointless activity, under this model, is not “natural”; it’s usually forced.
    • I’m sort of playing with the idea of a philosophy of “makerism”, in which the good guys are simply the people who do self-evidently useful things. Building a house or preparing a meal is obviously Useful Work. As is discovering a drug or inventing a tool. In makerism, if you’d have trouble explaining to a precocious twelve-year-old why you’re doing a useful thing, there’s a chance that what you’re doing is bullshit. I’ve sort of poked at the idea of measures of awesomeness and the ecosystem of industry before. The thing I’m trying to grope towards is productiveness. Not productivity, as in number of hours worked per day, or number of widgets produced per worker, but reaching towards usefulness, value, fruitfulness, substantialness, good-for-humans-ness.
  • My main update from job searching this time around (in mostly Silicon-Valley-based data science jobs) is that there is a thing called “fit” — how close the applicant’s background and skills are to what the employer is looking for — and the jobs you are an exact fit for will love you, and the ones you’re an imperfect fit for will reject you. For instance, it’s basically not worth it for me to even apply for jobs as a “data engineer”, because I’m not one. “Oh, it’s _close _to what I know and I can learn it on the job”? Nope. The right job is the one that’s dead center in the middle of your skillset.
    • Also, I had significantly better results applying to companies in the biomedical industry, I assume because I’ve done biomedical stuff in the past (systems-biology research in grad school, a personalized-medicine startup). The takeaway here is that I expect you have the best shot in jobs that correspond well to your entire background, including things that you might classify as a “side interest”. If you have a unique combination of skills, look for places that actively want that.
  • Bay Area software companies seem mostly pretty sane, in that they do not hire the flagrantly unqualified. Don’t expect to bluff your way in.
  • Because there are so many people sharing stories about the opposite experience, I think it behooves me to share mine; I didn’t experience anything that I’d classify as sexism during my job search, even though nearly all my interviewers were male, and so were nearly all the data scientists at the companies where I applied. The closest thing was being told that I was too “nervous” by one interviewer, which is sort of gendered in a statistical sense, but is also legitimately true of me, and not true of all women.
  • I have noticed that a fair number of companies are “segregated”, in that all the engineers are Asian (and foreign-born) while all the managers are white. It seems to correlate really well with, for lack of a better word, “lameness” — companies that are stagnant, hierarchical, complacent, don’t have a strong engineering culture, etc. I now consider racial glass ceilings to be a red flag.
  • Skills I wish I’d had: better memory for SQL syntax (yes, really), deep learning, computer vision, ETL pipelines
  • Skills I was glad I had: Spark, familiarity with the Python scientific computing & ML libraries, basic ML skills at the level of Hastie & Tibshirani, basic algorithms & data structures.
  • In technical interviews, a lot comes down to “fluency” or “execution” — can you solve simple math and programming problems correctly and quickly? are you checking for small errors? It’s very g-loaded, but I think there’s a skill of “turning your g on”, getting into “performance mode”, which I learned from years of being a math contest kid, and felt myself relearning as I went through the job search process. If you know what I’m talking about, focus on cultivating_ that_, through repetitive practice of fairly-easy things with a high bar for accuracy, rather than studying super-advanced topics.