Clojurians Log v2

what's wrong with starting with a superficial understanding and working backwards? isn't that how we all become experts anyway?

I guess that this is possible. And I think, that you enter a rabbit hole, where it is really hard to understand the code and what it does. What do you want to achieve? What do you mean with “control randomness”?

I am the master or rabbit holes unfortunately 😉 . I am the maintainer of https://probprog.github.io/anglican/ and I was thinking about replacing its more limited CPS transform compiler to use cloroutine. I agree that it is easy to create a mess with Lisp and meta-programming, but I do think the Clojure community would benefit from more directly exploring ideas from the PL community, for example Racket. Barely anybody from the PL world I talk to knows much about Clojure or would consider doing research in it. The pragmatism of Clojure is good to keep the ecosystem straight, but there is also not much radical happening for a Lisp. I was just checking barliman again, the Scheme community has a lot of good ideation. Anyway, this is a tangent...

I have some more crazy ideas in the direction of JIT compilation and online rewrites of the code for inference purposes (i.e. running your program in reverse or speeding up simulations/executions through approximations/shortcuts). In Clojure many people like logic programming a la Datalog (I also maintain Datahike), these ideas can to a degree also be applied to use programs with different variables assigned and being free. Probabilistic programming is doing this with weighted measures, which is arguably the most general setting.

I am trying to figure out whether one could build an OpenAI competitor on this set of ideas, but honestly it is hard to find people for that. They either only do deep learning or they have a superficial understanding of it, but are good programmers. I have been prototyping an AI engine that can do things like screen recording and automatic task execution https://simm.is/screen/491680819, and I am currently porting it to Electric and missionary, among other things.

(maybe that helps, maybe it just looks like rabbit holing)

woah, I was expecting something in the direction of derterministic programming. Really impressive stuff 🙂 Cloroutine might be a good fit for CPS. As far as I know missionary does not recompile clojure. Only cloroutine does. And cloroutine is only used in a few places in missionary. Namely ap, sp, and cp macros.

Right. A friend of mine worked on https://www.tweag.io/blog/2023-10-12-rhine-bayes/. They integrated FRP through monad transformers. I think Haskell is beautiful because in the end everything is defined just through a few algebraic expressions and structures (at least rhine bayes was beautiful to look at). I think Clojure could be more like this without enforcing a rigid type checker, e.g. by offering a wide range of standard protocols (type classes) for basic algebraic operations (monoids, groups etc.) and construction of algebraic data types. It might not fit the community well, but I think there is value in learning to think that way. I agree with Rich Hickey's specific takes on the Maybe monad in "maybe not", but I think the overall attitude towards algebraic thinking in the Clojure community limits its conciseness a bit. There is also considerable progress in this space, that is neglected under the notion that verification is not worth it.

likely a bug https://github.com/leonoel/missionary/issues/98

This pattern works

(m/ap
  (let [resource (m/?> (m/observe
                         (fn [!]
                           (! (<init-resource>))
                           #(<dispose-resource>))))]
    (<use-resource> resource)))