Whiteboard problems in pure Lambda Calculus (2017) | Hacker News Reader

Whiteboard problems in pure Lambda Calculus (2017)

jtolio.com

87 pointsjesperht26 comments

IMO Lambda calculus should be taught wider in CS programs. It brings useful models, builds (if one likes theory) good background to functional programming paradigms, etc.

However, to me, this post was hard to read. Wikipedia https://en.wikipedia.org/wiki/Lambda_calculus provides a much better introduction: from informal motivation to formal descriptions. My 2c.

I learned it in school, so I don't think it is that uncommon. Barely remember it though. Doesn't come up a ton.

I want to budget cognitive learning cycles to learning lambda calculus, but I've tried superficially a couple of times and couldn't find a practical reason to do so. Prioritizing picking up a more nuts & bolts skill has always seemed more engaging and lucrative. If I already know some Lisp-ish stuff from languages I already know, is there some benefit to picking up Lambda Calculus that isn't super obvious?

Just my 2c -- maybe not, this is just a way of thinking about problems that might come handy in some usually unexpected way. Also, it could be cool to mumble to your geek friends on use cases of alpha conversion and beta reduction.

Seriously though, I do not know Lisp, but if you know OCaml you probably have a good feel for lambda calculus just from practical application of the language. My 2c.

Learning OCaml is something I've wanted to do. I'll add this to the list of reasons to learn it :) Thanks!

> IMO Lambda calculus should be taught wider in CS programs. It brings useful models, builds (if one likes theory) good background to functional programming paradigms, etc.

I'm surprised than anyone thinks this. Unless you're in a very specialized area, it's about as useful as getting practice with writing algorithms using Turing Machines.

(Yes, it offers a very different perspective to the usual Von Neumann machine, but learning graph reduction as evalation is just as valuable, if not more.)

I stopped reading after...

(λU.(λY.(λvoid.(λ0.(λsucc.(λ+.(λ*.(λ1.(λ2.(λ3.(λ4.(λ5.(λ6.(λ7.(λ8.(λ9.(λ10.(λnum.(λtrue.(λfalse.(λif.(λnot.(λand.(λor.(λmake-pair.(λpair-first.(λpair-second.(λzero?.(λpred.(λ-.(λeq?.(λ/.(λ%.(λnil.(λnil?.(λcons.(λcar.(λcdr.(λdo2.(λdo3.(λdo4.(λfor.(λprint-byte.(λprint-list.(λprint-newline.(λzero-byte.(λitoa.(λfizzmsg.(λbuzzmsg.(λfizzbuzzmsg.(λfizzbuzz.(fizzbuzz (((num 1) 0) 1)) λn.((for n) λi.((do2 (((if (zero? ((% i) 3))) λ_.(((if (zero? ((% i) 5))) λ_.(print-list fizzbuzzmsg)) λ_.(print-list fizzmsg))) λ_.(((if (zero? ((% i) 5))) λ_.(print-list buzzmsg)) λ_.(print-list (itoa i))))) (print-newline nil)))) ((cons (((num 0) 7) 0)) ((cons (((num 1) 0) 5)) ((cons (((num 1) 2) 2)) ((cons (((num 1) 2) 2)) ((cons (((num 0) 9) 8)) ((cons (((num 1) 1) 7)) ((cons (((num 1) 2) 2)) ((cons (((num 1) 2) 2)) nil))))))))) ((cons (((num 0) 6) 6)) ((cons (((num 1) 1) 7)) ((cons (((num 1) 2) 2)) ((cons (((num 1) 2) 2)) nil))))) ((cons (((num 0) 7) 0)) ((cons (((num 1) 0) 5)) ((cons (((num 1) 2) 2)) ((cons (((num 1) 2) 2)) nil)))))

My team at Vivint, the Space Monkey group, stopped doing whiteboard interviews a while ago. We certainly used to do them, but we’ve transitioned to homework problems or actually just hiring a candidate as a short term contractor for a day or two to solve real work problems and see how that goes. Whiteboard interviews are kind of like Festivus but in a bad way: you get the feats of strength and then the airing of grievances. Unfortunately, modern programming is nothing like writing code in front of a roomful of strangers with only a whiteboard and a marker, so it’s probably not best to optimize for that.

The logical way forward is then to actually just hiring a candidate as a short term contractor for a week or two to solve real work problems on site, in a team with other employees and see how that goes.

The problem with that approach is that the company must commit. You can't embed 100 candidates into your team but you can easily send 100 identical homeworks and then only bother to analyze those that make it to the face to face interview. As a bonus, it optimizes for subservience, a perennial business resource.

An even bigger problem is that most good candidates have their current job and could be open to better opportunities, but would not consider going into this hung up state. Do you expect me to quit my job before doing the tryout? Even if not and, say, you like me, when do you expect me to quit my current job? I need at least a couple of months to cleanly exit without burning bridges and I am not going to even start this process until I have a firm offer from the new company.

Yeah. Because a programmer in a new team is likely to be productive after a week or two. /s

In my experience, good and bad candidates differentiate themselves pretty quickly. Being productive is not the expectation, it's about gathering more empirical data on the candidate's actual skills with real development.

Which is a really awful way of hiring. Forcing someone to go through the hassle of being a contractor for a couple of weeks, only so you can decide whether you want to hire them afterwards. I suggest you don't try this in Belgium, where employee protection is a thing and this sort of nonsense doesn't fly.

Unless, of course, you're looking for a long term contractor and not an employee.

What exactly is nonsense about it? They get paid for part-time work they otherwise would have to do for free during an interview or as homework.

Everything is nonsense about it. In most parts of the world, setting up to be a contractor (as opposed to an employee) is a hassle with various implications (tax, retirement, insurance,..). Doing it for a couple of weeks as part of a hiring process? Get real.

Getting someone to do actual work during an interview is also a no-no. As soon as someone walks through your door and you ask them to do actual work, they're your employee, period. They are now automatically entitled to compensation and a minimum severance period. You can interview people in almost any way you want, but as soon as it involves doing actual work, you're toast. This may be different where you live, but it strikes me as being unethical at the very least.

Hiring is hard. But exploiting people (by expecting them to do actual work, homework, forcing them into temporary contracting) is not the answer.

This is kind of getting off topic, but this would not at all work for me, and I suspect most job-seekers. When I am job-searching I don't put my search or my current job on hold for a couple weeks while a potential employer 'evaluates' me.

EDIT: this would be nice on the other hand to get to experience the 'real' culture and work environment at a place before I decide to commit to them.

Extra weirdness with this turing-complete combinatory logic with only a single combinator [1,2]. Combinatory logic being like lambda-calculus but where are only allowed to use some predefined functions and function application (you can't create your own functions). Bonus: since iota/jot/zot only have 1 symbol + application, programs are binary trees (with the combinator in the leaves) which we can write in prefix order as "1" for leaf and "0<program><program>" for node. And now (modulo some io details) we have a cool language where any bitstring is a valid program.

[1] http://www.nyu.edu/projects/barker/Iota/zot.html [2] https://en.wikipedia.org/wiki/Iota_and_Jot

Every day I feel amazed at how much smarter other people are.

Yes this sound unrealistic in most cases. Taking a leave from your current job to try out another one is very risky.

Lambda calculus is simple and easy to understand. I learnt it in university from scratch and I think that everybody can do it. Set theory has a quite similar, manageable, learning curve.

And set theory and lambda calculus have nice 'interoperability' (e.g. λx[Dog(x)] is a characteristic lambda function for the set {x | x is a dog}.)

Previously at https://news.ycombinator.com/item?id=13914004.

Factorials in pure lambda calculus, compiled to native code and running in non-geological time:

http://www.flownet.com/ron/lambda-calculus.html

Could anyone describe practical benefits to studying lambda calculus (other than, perhaps, to better understand the underlying concepts in LISP)? It seems kind of interesting because it is a way to describe what programs do with a very small number of building blocks. But why is it useful? What insights does it bring?

not a good student, can't give you a good overview but this kind of blew me away: https://en.wikipedia.org/wiki/Curry%E2%80%93Howard_correspon... - you also can explain the y-combinator from first principles. Not sure how practical that is :)

Reply on news.ycombinator.com