Carl Lange

@gwr Also not a mod anymore for at least 2 years… chat.stackexchange.com/transcript/2234?m=58451371#58451371

(Maybe they do, I haven't paid for 13.3 yet :P)

Actually I'm a bit surprised they don't hide away the actual API cost behind Wolfram AICredits™ or something weird like that.

@ChrisK I've been using ChatGPT's API quite heavily for the last few months - it writes about 70% of my non-WL code these days - and I've yet to spend more than €10 per month. I'm pretty curious how much API usage 13.3's new features require and how much it'll end up costing.

To my inexpert eye it doesn't seem like this is a new conceptual algorithm (like quicksort or bubblesort) but more of a mechanical improvement to the existing sort.

@Nasser If I'm not mistaken, this algorithm has already been in LLVM for some time (April 2022 - reviews.llvm.org/D118029). I'm not sure how much of Mathematica is compiled with LLVM, if any, but if so it might already be in there.

@kirma I can't believe I've been using the geo functionality for years without seeing this

@kirma Honestly, GeoMidpoint has been missing from the language for years!

I agree though, questions like "How can I generate this code with ChatGPT' or "how do I solve this equation in natural language" will want to be off-topic. The harder part is what happens when people ask ChatGPT to write code to do X and then don't understand what's wrong in the resulting code, and how to help those people beyond RTFM

@b3m2a1 I think it will become fairly hard to distinguish some of these questions, though - it's definitely hard to determine whether code has been generated by ChatGPT or by a human for certain languages and environments. At the minute I think WL doesn't have this problem because ChatGPT doesn't have quite enough training data, but with an effort from Wolfram we're definitely going to get there.

@Nasser Yes - (un)fortunately I'm fairly sure that's what all programming-specific forums will become in the next couple of years as ChatGPT functionality gets integrated into different environments. I am a mediocre lisp programmer but I've been able to write thousands of lines of lisp with ChatGPT. However I've kept any mistakes to myself and not bothered SO about it - others will be less scrupulous, I'm sure :)

I've build some integrations between ChatGPT and my other programming tools and I am absolutely faster at my job (and having more fun) because of it - having similar things in Mathematica would be amazing.

@Feeds Oooh, this one's exciting. I am a lot more interested in LLMs coming to Mathematica than I was about all the blockchain stuff. If 14 has a chatgpt powered natural language interface I will definitely upgrade.

@kirma Yeah, this drives me nuts. I don't think it uses any credits but there's no reason for this to be server-side. mathematica.stackexchange.com/questions/195482/…

@kirma For years now, yes - I remember asking for it on the SW livestreams in like 2019. Maybe I should write a function repository function to call ogr2ogr...

@kirma Yes, I really wish they would have proper support for EPSG codes built-in, it drives me nuts that I have to write a bespoke converter or use ogr2ogr every time I come across something in a new projection

@kirma Oh yes, it is code from your answer, I hope you don't mind that I used it in that example. It was just the first Q/A I came across that looked like a good candidate!

@kirma Oh, totally, it wasn't a great example. To be honest it's not great at ascii art as output, but it does do a great job of parsing it as input:

@VitaliyKaurov 3.5, using my emacs package: github.com/CarlQLange/chatgpt-arcana.el

@VitaliyKaurov ChatGPT can do it, kinda :)

@Musang This functionality exists in emacs at least, with a package called expand-region: github.com/magnars/expand-region.el

@kirma Honestly my favourite thing to do with Mathematica is this kind of messing about. So many things it would have been fairly hard and time-consuming to calculate 30 years ago are just a couple of functions away.

@kirma That's super cool, I love it!

@Feeds This is really cool. I've spent a little while writing an Emacs package for ChatGPT but the whole time I was wondering what it would look like in Mathematica. Can't wait to see where this goes.

Or, it'll turn into self-driving cars and we'll never really see it perfectly executed. Who knows. Exciting times, though.

Have had a few thoughts about feeding the WL documentation into eg gpt-index.readthedocs.io or into some magic LangChain stuff (langchain.readthedocs.io). It's really fun at the moment, because it's such early stages into the applications of these things, and what it might be like in three years is really exciting.

@Feeds I've been playing around a lot with ChatGPT and other new-age models - Stable Diffusion etc - and I honestly can't wait to see if and how they get integrated into Mathematica. I never used all the crypto functions that were added in that heyday, but I'm really hoping for a similar amount of support offered to LLMs and generative AI in general.

@kirma Is this what their mugs and so on come in? I have often thought about buying a mug or similar, but the shipping costs to Ireland were more than the item cost... this would explain why :D

@b3m2a1 Yes, it's a shame that it's hard to convince people that it really can be great. I suppose I would have the same feelings if someone tried to convince me that R is genuinely sensible to use outside of very small use cases.

@user21 Yes, often the price is the sticking point, although I think it might be more a question of: they know what they write in python is allowed to be used and distributed however they like, be it open source, GPL, whatever. They don't know that about Mathematica and I guess it is hard to convince them that they should try it. I suppose they want to build their houses on the same foundations that everyone does

@gdelfino Ah, that's cool! I watch Formula E a lot and know that ZF works with at least one of the manufacturers. I've always felt that Mathematica would be very useful in autosport in particular ;)

In any case pretty much all my clients are small tech-focused companies who have teams who don't know Mathematica so that compounds the issue. I dream of working for a research institute or similar where it doesn't really matter what tech I'm using.

I find it tough sometimes because I find Mathematica 100x more efficient than any other programming environment I use, and then switching back to python or whatever for client work feels like running through treacle.

@tush I freelance, but don't use Mathematica for any work-related stuff - the license is hard to sell to clients. I might sometimes prototype something in Mathematica and then actually write it in Python or whatever to deliver to the client. I wish I could use Mathematica in the workplace but I haven't yet found a client who's happy with the licensing arrangements.

writings.stephenwolfram.com/2023/01/…

@kirma Even more surprising, it generates passable Irish! How there was enough in its training data I will never know.

To me it feels a lot like pair programming with a junior developer.

(adventofcode.com/2022 for the uninitiated)

Has anyone been doing Advent of Code in WL? I've been doing it this year to try and get back into Mathematica and finding it very enjoyable. I don't like my solution for today's puzzle (day 7) and was curious if anyone had an elegant solution. I'm sure it's doable with some nice FoldList incantation...

Have you seen URLShorten by the way? It's not as short as the goo.gl links but at least thematically appropriate ;)

@andre314 I have never understood this until now, it's always confused me!

Very nice!

@Szabolcs Works for me on Linux

(Though I eventually rewrote the code in another language for performance and portability reasons)

I forgot to mention it while the exhibition was on, but if you happened to go to Walk! in the Schirn Kunsthalle in Frankfurt, there was some Mathematica-generated artwork schirn.de/en/exhibitions/2022/walk ellieberry.com/blog/tag/fractured+landscapes

I've seen WL mentioned more and more on Hacker News lately... getting worried that my secret productivity weapon is going to be more well-known