3 ms·

I love theory, and I also really like lambda calculus, but I feel like this sentiment applies to most theory, particularly stuff after undergrad. I spent a not

I love theory, and I also really like lambda calculus, but I feel like this sentiment applies to most theory, particularly stuff after undergrad. I spent a not-insignificant amount of time learning how to do proofs with Isabelle. I learned a lot about inductive proofs, set theory, meta-logic, and challenged myself to prove a lot of the stuff I had previous taken for granted (e.g. proving that different sorts refine each other). I enjoyed it, and similarly was convinced that this was going to be some life-changing thing that changes my career trajectory and... Nothing changed. No one in charge of companies gives a shit about theory. They all claim that they love theory, they claim that they are very research focused, they claim that they value all the time you spent learning this stuff, but in reality they really just want you to change the color of buttons, or change the format of dates, or add a field to a JSON. It sometimes feels like no software engineer but me actually wants to learn any math, and will refuse to touch anything even resembling it. And I'm not picking on Isabelle here; I've had similar results trying to pitch TLA+ and Coq and Agda for some of the more error-prone parts of the codebase, with different sales-pitches, and without fail the managers will always say that they "will look into it", and promptly do absolutely nothing. The first two times a manager said that, I believed them, but after that I realized that they're just trying to shut me up and tell me "no" politely. It was enough to depress me, and it still kind of does. I still think learning stuff for fun is worth it, but I'd be lying if I told you if I knew why.

hnoyrnndd 3y ago
A friend of mine worked at Rockwell Collins and wrote proofs, and theorem provers, all day. If you work on webshit like me you won’t get to use these skills much. (On the other hand, recursing over the structure of JSON-like data often feels like 80% of the job, so I think the skills come into play at least a tiny bit).
- tombert 3y ago
  I work on web-adjacent-shit, mostly data processing stuff in Kafka that eventually ends up on the web. I don't get to recurse over JSON, occasionally I get to design a system from scratch and that's more fun, but it's usually not more complicated than "draw boxes that point to other boxes and/or cylinders on screen". Sometimes I draw a picture of cloud. I like my job just fine, it's a decent job, and I like my managers and coworkers as well, but it's just disappointing that enthusiasm for math and theory is what got me to this stage of my career, but I never really got to use it, and I don't see that ever really changing for me. Maybe I will write a paper at some point at least.
  - andoando 3y ago
    Yeah I am at the same place. Considering going for a phd because at this point I cant muster any kind of interest in corporate tech.
    - tombert 3y ago
      From experience, that has its own share of frustrations as well. You get to work in theory and that’s fun, but if you decide to do a PhD while also working a desk job, I recommend you do not tell your employers. I made the mistake of telling an employer, and whenever I made any mistake in my work, no matter how small, the employer would immediately say that the PhD work is distracting me, and my focus not BigCo. My PhD is on an indefinite hiatus right now, because it’s something I am questioning the utility of right now.
      - andoando 3y ago
        That sucks. I cant imagine trying to work a job and do a phd at the dame time. That would take out any joy of it for me. I can definitely see a lot of utility depending on what you do research in. I bet a lot of ML doctoretes are making big bucks, but that's relatively niche.
        tombert 3y ago
        The PhD I was doing was in formal methods, specifically in regards to timing semantics. I think it’s super neat, but sadly I am in a minority. ML is cool but I was never able to get super into it. I use ChatGPT but I never had a ton of desire to get into the guts of it. I was always more interested in discrete math and formal logic.
        andoando 3y ago
        That sounds interesting, can you tell me a bit more? What does formal methods in regards to timing semantics mean? Im in the same boat. I like abstract logic/math and ML never caught much of my interest. I want to start applying but Im not even sure how to find departments that would fit what Im looking for. Maybe I need to take this more seriously and start reading papers and self study.
andoando 3y ago
But now you can write your own json parser
jvanderbot 3y ago
I use the deep theory experience a ton when designing software or systems. But I don't do an actual proof hardly ever. Being able to reason about a problem and intuit the extent of what's technologically possible to solve is insanely useful. But in most places that I work everyone around me can do that too, so we're constantly battling program managers and C suite guys / gals who just don't see why everything is so slow, difficult, un-scalable, unpredictable, etc. They want a stupid thing that works half the time, and they'll just turn the camera off the other half of the time.
- catgary 3y ago
  Yeah, I’m currently an ML scientist but I regularly run into shit where I have no clue how you’d solve the problem without having a compilers/functional programming background.
eddd-ddde 3y ago
I did a lot of math before I began learning programming. This was my experience: Recursion? Oh you mean proof by induction? Cryptography? Oh you mean number theory? Neural networks? Oh you mean calculus? Almost everything I learn while programming can be associated with some theory I learnt before for math.
- tombert 3y ago
  That's kind of what annoys me though; it's almost like the word "math" is toxic to people. When I try explaining a correspondence between something that they're doing via code, and how maybe learning a bit of the mathematics behind it might be useful, people kind of just roll their eyes, acting like I asked them to go get three PhDs, and acting like it's just one of those weird "Tombert things". I don't know anything about neural networks yet (though I really need to get on that), but I have noticed the other two examples you mentioned as well; recursion is more or less applied inductive proofs, a lot of crypto boils to number theory. My dream is to some day convince a manager to give me budget to spend a few weeks designing a new system and proving correctness with TLA+. I'm not saying it's terribly likely, but a man can dream.
  - keybored 3y ago
    An unwelcome comparison: For certain people you can’t just teach Buddhist meditation using Buddhist terminology and context because that’s “woo”. However if you dress it up in vaguely “neuroscience” terms then it’s fine. And preferably you have some scientific studies at hand to prove that meditation has an effect. (Although no one demands biomechanic studies from their intro tennis teacher.) Compare with all the moaning and complaining that comes from some people when you talk about functional programming and how it might relate to computer science.[1] Then you might have a better chance inventing ad hoc words for basics like “map” (preferably very prosey) and claiming that Martin Fowler invented it. [1] This goes triply for anything having to do with proofs, at least proof software associated with FP.
    - richk449 3y ago
      Interesting point. One difference: It is easy to tell if a tennis instructor is giving “good” instruction, where good means instructions that help you be a better tennis player. You just look at their students. If you want to know if they give good fundamentals for long term growth, you look at their much older students. How do you know if a meditation teacher is giving good instructions? Seems much harder to me. It is interesting that many of the traditional religions have hard to quantify benefits. Buddhism offers a personal inward help, and the abrahamic religions often benefit communities.

hnoyrnndd 3y ago

A friend of mine worked at Rockwell Collins and wrote proofs, and theorem provers, all day. If you work on webshit like me you won’t get to use these skills much. (On the other hand, recursing over the structure of JSON-like data often feels like 80% of the job, so I think the skills come into play at least a tiny bit).

tombert 3y ago
I work on web-adjacent-shit, mostly data processing stuff in Kafka that eventually ends up on the web. I don't get to recurse over JSON, occasionally I get to design a system from scratch and that's more fun, but it's usually not more complicated than "draw boxes that point to other boxes and/or cylinders on screen". Sometimes I draw a picture of cloud. I like my job just fine, it's a decent job, and I like my managers and coworkers as well, but it's just disappointing that enthusiasm for math and theory is what got me to this stage of my career, but I never really got to use it, and I don't see that ever really changing for me. Maybe I will write a paper at some point at least.
- andoando 3y ago
  Yeah I am at the same place. Considering going for a phd because at this point I cant muster any kind of interest in corporate tech.
  - tombert 3y ago
    From experience, that has its own share of frustrations as well. You get to work in theory and that’s fun, but if you decide to do a PhD while also working a desk job, I recommend you do not tell your employers. I made the mistake of telling an employer, and whenever I made any mistake in my work, no matter how small, the employer would immediately say that the PhD work is distracting me, and my focus not BigCo. My PhD is on an indefinite hiatus right now, because it’s something I am questioning the utility of right now.
    - andoando 3y ago
      That sucks. I cant imagine trying to work a job and do a phd at the dame time. That would take out any joy of it for me. I can definitely see a lot of utility depending on what you do research in. I bet a lot of ML doctoretes are making big bucks, but that's relatively niche.
      - tombert 3y ago
        The PhD I was doing was in formal methods, specifically in regards to timing semantics. I think it’s super neat, but sadly I am in a minority. ML is cool but I was never able to get super into it. I use ChatGPT but I never had a ton of desire to get into the guts of it. I was always more interested in discrete math and formal logic.
        andoando 3y ago
        That sounds interesting, can you tell me a bit more? What does formal methods in regards to timing semantics mean? Im in the same boat. I like abstract logic/math and ML never caught much of my interest. I want to start applying but Im not even sure how to find departments that would fit what Im looking for. Maybe I need to take this more seriously and start reading papers and self study.

andoando 3y ago

But now you can write your own json parser

jvanderbot 3y ago

I use the deep theory experience a ton when designing software or systems. But I don't do an actual proof hardly ever. Being able to reason about a problem and intuit the extent of what's technologically possible to solve is insanely useful. But in most places that I work everyone around me can do that too, so we're constantly battling program managers and C suite guys / gals who just don't see why everything is so slow, difficult, un-scalable, unpredictable, etc. They want a stupid thing that works half the time, and they'll just turn the camera off the other half of the time.

catgary 3y ago
Yeah, I’m currently an ML scientist but I regularly run into shit where I have no clue how you’d solve the problem without having a compilers/functional programming background.

eddd-ddde 3y ago

I did a lot of math before I began learning programming. This was my experience: Recursion? Oh you mean proof by induction? Cryptography? Oh you mean number theory? Neural networks? Oh you mean calculus? Almost everything I learn while programming can be associated with some theory I learnt before for math.

tombert 3y ago
That's kind of what annoys me though; it's almost like the word "math" is toxic to people. When I try explaining a correspondence between something that they're doing via code, and how maybe learning a bit of the mathematics behind it might be useful, people kind of just roll their eyes, acting like I asked them to go get three PhDs, and acting like it's just one of those weird "Tombert things". I don't know anything about neural networks yet (though I really need to get on that), but I have noticed the other two examples you mentioned as well; recursion is more or less applied inductive proofs, a lot of crypto boils to number theory. My dream is to some day convince a manager to give me budget to spend a few weeks designing a new system and proving correctness with TLA+. I'm not saying it's terribly likely, but a man can dream.
- keybored 3y ago
  An unwelcome comparison: For certain people you can’t just teach Buddhist meditation using Buddhist terminology and context because that’s “woo”. However if you dress it up in vaguely “neuroscience” terms then it’s fine. And preferably you have some scientific studies at hand to prove that meditation has an effect. (Although no one demands biomechanic studies from their intro tennis teacher.) Compare with all the moaning and complaining that comes from some people when you talk about functional programming and how it might relate to computer science.[1] Then you might have a better chance inventing ad hoc words for basics like “map” (preferably very prosey) and claiming that Martin Fowler invented it. [1] This goes triply for anything having to do with proofs, at least proof software associated with FP.
  - richk449 3y ago
    Interesting point. One difference: It is easy to tell if a tennis instructor is giving “good” instruction, where good means instructions that help you be a better tennis player. You just look at their students. If you want to know if they give good fundamentals for long term growth, you look at their much older students. How do you know if a meditation teacher is giving good instructions? Seems much harder to me. It is interesting that many of the traditional religions have hard to quantify benefits. Buddhism offers a personal inward help, and the abrahamic religions often benefit communities.