Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I'm not mutable, I'm partially instantiated

blog.dnmfarrell.com

Post reply on HN
Print view View original

11–20 of 77 posts

Re: I'm not mutable, I'm partially instantiated

#11

Post by sixfiveotwo »

weinzierl wrote: post #7
lelag wrote: post #2

I've noticed quite a lot of Prolog content lately. I don't know if it's just a case of the Baader–Meinhof phenomenon, but it looks to me that recent article[1] about Prolog being able to improve LLM reasoning abilities renewed some interest in the language. [1] https://shchegrikovich.substack.com/p/use-prolog-to-improve-...

Which is strange considering how old and limited Prolog in a sense is. I wonder why no superset ever gained traction and what it would look like. I imagine it fitting somewhere in the hierarchy Theorem Prover ⊃ SMT Solver ⊃ SAT Solver since Theorem Prover ⊃ Prolog

Indeed, you can get a lot more from dependent types than Damas-Hindley-Milner inference, yet does it mean that you should use the former everywhere?

Re: I'm not mutable, I'm partially instantiated

#12

Post by Epa095 »

weinzierl wrote: post #7
lelag wrote: post #2

I've noticed quite a lot of Prolog content lately. I don't know if it's just a case of the Baader–Meinhof phenomenon, but it looks to me that recent article[1] about Prolog being able to improve LLM reasoning abilities renewed some interest in the language. [1] https://shchegrikovich.substack.com/p/use-prolog-to-improve-...

Which is strange considering how old and limited Prolog in a sense is. I wonder why no superset ever gained traction and what it would look like. I imagine it fitting somewhere in the hierarchy Theorem Prover ⊃ SMT Solver ⊃ SAT Solver since Theorem Prover ⊃ Prolog

The post in OP has the following hyphothesis:

> Why Prolog? [...] Due to it's declarative nature it might be a bit easier for LLMs to generate code, because LLM doesn't need to generate precise control flow.

This is an interesting point, and my guess is that prolog is the declarative programming language with most example code out there for it to learn on(excluding SQL). Alternatively it could try to create some problem description for an automated theorem prover. My (absolute ignorant) guess is that the prolog aproach works better for two reasons:

- The amount of prolog code in the training set is higher

- It is still only able to create code for problems easy enough that prolog can handle them.

Re: I'm not mutable, I'm partially instantiated

#13

Post by Epa095 »

knome wrote: post #10
weinzierl wrote: post #7

Earlier quoted context omitted.

Which is strange considering how old and limited Prolog in a sense is. I wonder why no superset ever gained traction and what it would look like. I imagine it fitting somewhere in the hierarchy Theorem Prover ⊃ SMT Solver ⊃ SAT Solver since Theorem Prover ⊃ Prolog

limited in what sense? prolog is turing complete.

He might be thinking about the SLDNF resolution happening. But yes, you can implement any prover in prolog. This distinction is discussed a bit here: https://www.metalevel.at/prolog/theoremproving

Re: I'm not mutable, I'm partially instantiated

#14

Post by dsabanin »

weinzierl wrote: post #7
lelag wrote: post #2

I've noticed quite a lot of Prolog content lately. I don't know if it's just a case of the Baader–Meinhof phenomenon, but it looks to me that recent article[1] about Prolog being able to improve LLM reasoning abilities renewed some interest in the language. [1] https://shchegrikovich.substack.com/p/use-prolog-to-improve-...

Which is strange considering how old and limited Prolog in a sense is. I wonder why no superset ever gained traction and what it would look like. I imagine it fitting somewhere in the hierarchy Theorem Prover ⊃ SMT Solver ⊃ SAT Solver since Theorem Prover ⊃ Prolog

Another attractive feature of Prolog is that it’s homoiconic, like lisp.

Re: I'm not mutable, I'm partially instantiated

#17

Post by openasocket »

I've never used it in production, but I have a deep love of Prolog, just because of how different it is from any other programming language I've used. As a programming paradigm, it found it as eye-opening as functional programming. What I found interesting is that you are operating on logical statements and pattern matching, which often means that the same "function" can be used for multiple different things. For example:

append([], List, List).

append([Head|Tail], List, [Head|Rest]) :- append(Tail, List, Rest).

Is a simple list append function: append(X, Y, Z) is a predicate that match if Z is the list of all elements of X, followed by all elements of Y. You can use it to concatenate lists, of course. But you can also use it to confirm that a particular list start with a particular sequence, or ends with a particular sequence! The idea that the same predicate can be used in multiple different ways is really fascinating to me. I have no idea to what extent that would scale in a large system, but it's very interesting

Re: I'm not mutable, I'm partially instantiated

#19

Post by agumonkey »

openasocket wrote: post #17

I've never used it in production, but I have a deep love of Prolog, just because of how different it is from any other programming language I've used. As a programming paradigm, it found it as eye-opening as functional programming. What I found interesting is that you are operating on logical statements and pattern matching, which often means that the same "function" can be used for multiple different things. For exa…

Same, I love the bidirectional aspect of relations

Re: I'm not mutable, I'm partially instantiated

#20

Post by wizzwizz4 »

openasocket wrote: post #17

I've never used it in production, but I have a deep love of Prolog, just because of how different it is from any other programming language I've used. As a programming paradigm, it found it as eye-opening as functional programming. What I found interesting is that you are operating on logical statements and pattern matching, which often means that the same "function" can be used for multiple different things. For exa…

It doesn't scale that well because integers are their own, opaque thing in Prolog, and the is predicate is unidirectional. However, there's no inherent reason this has to be the case: you can construct your own representation of the integers from the Peano axioms, and recover bidirectionality of addition. (You'll want to be using some typed variant of Prolog with the "unary tree of units" optimisation, otherwise integers take O(n) space in memory, but it's possible in the language.) Prolog works how it does, because it's (mostly) backwards-compatible with a facts database used for symbolic AI research, where all atoms were opaque.
Post reply on HN