07:41 <greenbagels> https://caml.inria.fr/pub/old_caml_site/caml-list/0838.html

07:42 <greenbagels> what does the comment " h is not generalized, since its definition is a function composition" mean?

08:12 <araujo> hello, how good is ocaml for developing games?

08:57 <dh`> it is possible, it is probably not the first choice

09:00 <octachron> greenbagels, this is the value restriction, only syntactic values can have generalized/polymorphic types (except when the relaxed value restriction triggers).

09:06 <greenbagels> octachron: i will read about this; thanks!

09:06 <greenbagels> the ocaml manual is suprisingly more readable than i thought it was when i first looked at it

10:34 <reynir> @leviroth thanks! type outer := t is excellent

19:14 <darchitect> hey guys, what do you think the future of PL research is if all we do is write tests for the language model to implement ?

19:15 <darchitect> do we need to consider what makes a good language for models to use as opposed to humans maybe.. sorry if this is to naive of a question I'm pretty new to the field

19:17 <companion_cube> we don't, LLMs won't replace programmers anyway :)

19:19 <darchitect> yeah but they will replace most of them. Why would you need a web dev if the models can implement a project the business person will be able to see if they like it or not, if they don't they can just regenerate a new project from scratch or tweak the current one ... maybe they won't replace them for systems programming, but that's like 1 % of devs out there

19:20 <companion_cube> don't drink the koolaid too much :)

19:21 <reynir> what I've seen generated using LLMs by people who don't know OCaml has been useless garbage that doesn't compile at all

19:21 <octachron> You mean the website where the LLM will have hallucinated the database, the authentification and the payment connector?

19:22 <darchitect> I've built a website for my dad who's a lawyer in 10 minutes, granted it's simple software, but before I wouldn't have even started doing the project because I knew it would take too much time and I will have to understand JS and web dev (which I personally find pretty boring)

19:22 <discocaml> <hockletock> ah but the business person won't realize that his payments are being hallucinated until after he's fired his devs

19:23 <darchitect> that's probably more than 100x improvement, because even if I had started I would be bored out of my mind googling for silly exceptions

19:25 <darchitect> I don't really believe programmers will be replaced, but I believe programming will become really easy for business people to learn it the way an experienced programmer can read a book on the business domain and be "good enough" to write software for it if the business domain is simple enough like most web dev stuff

19:25 <darchitect> by that time most currently employed developers will have to learn something else right ?

19:27 <darchitect> I fully understand there will be domains where you really need specificity in expressing intent, but this is not most domains

19:29 <reynir> darchitect: try see if you can make it solve something like this in OCaml https://adventofcode.com/2023/day/1

19:30 <darchitect> probably not, because this requires reasoning and we're not there yet

19:31 <darchitect> I would want to see where Google goes with reinforcement learning backend + LLM frontend kind of an architecture

20:10 <discocaml> <ulugbek3209> does `opam update` get stuck for anyone else on macos arm64?

22:38 <discocaml> <darrenldl> darchitect: i think that speaks more about the sample size for the particular kind of application you are developing for business people's needs than anything else

22:48 <darchitect> discocaml: that's true, at least based off of current cutting edge...probably for the better, don't really want to try to debug model weights :D

22:49 <discocaml> <Kali> reynir: bing chat could solve day 2 part 1 with only mild human intervention (to remove the footnotes that accidentally got into the code)

22:49 <discocaml> <Kali> no assistance was otherwise given, just asking it to read the problem, write a summary, then write a solution in ocaml

22:51 <discocaml> <Kali> and it worked first try

22:54 <discocaml> <Kali> of course, that's cherry-picking: it couldn't solve day 1 part 1 correctly within 4 messages of assistance

22:55 <discocaml> <Kali> LLMs are not going to replace programmers anytime soon

22:56 <discocaml> <Kali> if ever