[00:10] Funny that people complain about the image on the blog which is there only for decoration [00:13] I agree with general sentiment that llms are stochastic parrots, but I disagree about `no never use them` sentiment [00:16] it's a bit like with code templates or generators, use too much and your technology becomes too obscure too maintain [00:16] use not enough and you're writing assembly code [00:16] by hand [01:08] 🤔 [01:36] *** erooke left [01:50] *** erooke joined [01:58] *** hulk joined [01:59] *** kylese left [02:15] *** hulk left [02:15] *** kylese joined [03:23] *** acidsys left [03:24] *** acidsys joined [03:51] *** erooke left [04:07] *** erooke joined [04:17] *** ShimmerFairy left [04:19] *** ShimmerFairy joined [05:28] *** Sgeo left [06:07] *** erooke left [06:12] *** xinming left [06:13] *** xinming joined [06:22] *** erooke joined [08:00] *** rnddim joined [08:01] *** rnddim left [08:01] *** rnddim joined [08:02] *** ShimmerFairy left [08:03] *** rnddim left [08:03] *** rnddim joined [08:06] *** dakkar joined [08:09] *** rnddim left [08:10] *** rnddim joined [08:12] *** dakkar left [08:28] *** dakkar joined [08:30] This is an interesting analogy. Maybe it should be developed in more detail. [08:31] BTW, I am using templates with some of my LLM solutions. They provide faster and (far) more robust code generation. [08:31] *** rnddim is now known as ShimmerFairy [09:09] weekly: https://github.com/rakudo/rakudo/pull/6847 [09:09] lizmat, Noted! (weekly) [10:15] *** dakkar left [10:30] *** dakkar joined [10:41] easier reading: https://github.com/rakudo/rakudo/blob/682876f06bee8351c18a48475796888085336450/AI-POLICY.md [11:24] lizmat: any concern for ai assisted module submissions? should it be mentioned? [11:25] I think we can only enforce such a thing with the raku community modules [11:26] module authors are responsible for their modules [11:26] ok [11:26] of course, if a module is marked as LLM or non-LLM, could be grounds for users later on deciding whether or not to use such a modules [12:43] LLM / AI-agent assistance can be mentioned or described in the README. I also put a file "prompts.md" in the "./docs" folder in order to show (or trace) for which parts of the package LLMs were used. [12:45] My main goal with these kind of "prompts.md" files is to make the development more clear and reproducible. [12:46] But, obviously, they indicate are LLMs used or not. [12:47] Also, one can ask LLMs or AI-agents to analyze a package repository and determine to what degree that package was made via LLMs or AI-agents. [12:50] In Dutch we would say: We of WC-eend approve of WC-eend [12:52] I am not sure what do you mean. LLMs are very willing to rat on each other. (So far.) [13:07] WC-eend is a type of toilet cleaning fluid... and one of their (tongue in cheek commercials that every Dutch person of a certain age knows) they showed some guys in white lab coats with the WC-eend insignia, that they approved of WC-eend [13:27] *** dakkar left [13:32] *** vasko453558616 left [13:33] *** vasko453558616 joined [13:43] *** dakkar joined [14:05] *** dakkar left [14:21] *** dakkar joined [14:51] *** fondue left [14:51] *** scott2 joined [15:31] antononcube: can you name one of yr modules with the AI comments? [15:50] "Math::NIntegrate" is package I actively work one: https://github.com/antononcube/Raku-Math-NIntegrate/ . In its folder "./docs" see the files "Development-methodology.md" and "prompts.md". [15:50] thnx [15:58] TBH, LLMs were not as good with reprogramming Fortran code as I expected. I had to read the Fortran codes to verify the translation. 🙂 [15:58] But the LLM / AI-agent I used definitely sped up that translation. [15:58] my use is not so sophisticated. i just have "hal" do the grunt work and ensure i have tests for all. [15:59] For making tests and benchmarks LLMs are usually very good. [15:59] it sure helps when one's memory is fading... [16:00] and it is wonderful for **slow** typists [16:02] I used an LLM to convert one of the larger well known numerical integration suite to JSON. Then I wrote a Raku script to convert that JSON to Raku tests. (Reproducible research, blah, blah.) But, in general, I write the tests myself since I want to know how convenient is the design of the framework. [16:05] i sometimes have to fix the tests because it gets confused and uses some python or perl bits [16:09] You have to use proper language prompts. See for example use LLM::Prompts; llm-prompt('CodeWriterX')('Raku'). [16:22] *** dakkar left [16:31] *** stanrifkin joined [16:33] hm, i don't use that approach at all. i start with something like: create a raku module that [...describe reqs in detail...]. i have previously described my raku coding style, module construction style, etc. we go back and forth tweaking things. as it gets more complex i will zip the current code on my side and pass it to ai. ai will fix and zip it up an pass it back to me. i carefully keep stages in git for recovery if need be. [16:34] Here is an example how I use that "@CodeWriterX|Raku" prompt: [16:34] https://cdn.discordapp.com/attachments/633753286209699870/1559605331267424489/Screenshot_2026-10-13_at_12.33.37_PM.png?ex=6acfaf27&is=6ace5da7&hm=a8ae222387cfe034949c29510e1ced681e328ce893c9fafaf423c57e3604a90e& [16:36] Instead of using a Juputer magic cell that can be done with: llm-synthesize([ llm-prompt('CodeWriterX')('Raku'), , ], e => llm-configuration('ollama')) [16:37] i feel like ai has regressed a bit in remembering procs since we started [16:37] But if you are using an LLM agent you can just put those plain-English instructions and some instructions to just give Raku code. [16:38] No, I do not think so. 🙂 But the main use cases have changed. [16:42] *** sftp joined [16:42] *** sftp left [16:42] *** sftp joined [16:52] from what i saw of your LLM i wasn't sure how you used it. were instructions partly done in # comments? [16:52] In that screenshot above? [16:59] i missed that one. but is that code you pass as a text int ai's "slot" and it writes its response in its "view port"? [16:59] Yes. [17:01] if so that's essentially what i do but i haven't tried to keep those chunks in a catalog [17:01] <[Coke]> "This content is no longer available." [17:01] <[Coke]> (cdn.discordapp.com url) [17:05] *** z-raku joined [17:08] antononcube: so i guess we're both doing similar usage in the raku TIMTOWTDI way [17:09] I'd like to install rakudo star on my Windows work laptop, but I don't have admin and msi install is restricted. I have installed rakubrew and the latest rakudo (moar-2026.09). Can I use rakubrew to install the rakudo star equivalent packages? I can take this to raku-beginner if it is better there. [17:10] patrickb might know [17:10] can you get WSL without admin? [17:11] I think that is how the rakubrew install succeeded. [17:13] I'm sorry, I misunderstood. I'll check if I can get WSL working. [17:15] Looks like WSL is installing without complaint. [17:16] Nope, admin required. [17:19] Looks like I'll use zef to get the modules. [17:24] *** stanrifkin left [17:30] *** z-raku left [17:33] *** z-raku joined [17:42] *** z-raku left [17:58] *** [Coke] left [18:04] *** [Coke] joined [18:05] *** hankache joined [18:05] hello #raku [18:14] hankache! :-) [18:14] hey lizmat how are you? [18:14] long time... [18:15] heh... fine mostly [18:15] indeed [18:15] hope you're doing well as well, assuming so :-) nice work on https://raku.guide ± [18:15] ! [18:16] *** z-raku joined [18:17] Glad you're doing mostly well. Same here. [18:18] Thanks for the responses. 360.zef.pm is blocked from my org (I've requested it be reviewed to allow). There are some modules not found, e.g. File-Find, Terminal-ANSIColor,... I'll be grabbing them manually unless there is a source where zef can be pointed. [18:21] Thanks! After a long break, it feels great to start dipping my toes back in. [18:26] quit :off to install the packages [18:26] *** z-raku left [18:41] <[Coke]> You can install directly from github [18:42] <[Coke]> but you'd need to know where to look (and normally zef would use the url you mentioned to look it up) - you could look up a package on raku.land, and likely get the source URL from there. [18:43] makes me wonder whether zef should have a "raku.land" option to obtain the repo URL from there [18:46] *** coleman left [18:48] *** coleman joined [18:52] <[Coke]> I wouldn't want to have to make the fallback supported enough to be relied on. Having worked at companies that lock down the internet, getting them to review/open the site is probably the best option. [18:55] perhaps better to mirror the output of 360.zef.pm to maybe something like codeberg.org regularly [18:55] once a day or so [18:55] and allow zef to fall back to that if 360.zef.pm cannot be reached [18:57] *** coleman left [18:58] *** coleman joined [19:45] <[Coke]> What browsers do we want to support on raku.org. Chrome, Firefox, Safari, anything else that has a different engine? [19:46] <[Coke]> (Trying to review a web site PR and we don't have the infra to automatically check the render of a site on various browsers) [19:50] Those would be the core ones, yeah. Almost every browser is using their core renderers. [19:50] I think most mainstream ones are based on Chromium and Gecko, and Webkit on Apple ecosystem [19:50] The only thing I can think of is making sure it works on both desktop and mobile browsers [19:51] I have tested locally on chrome and firefox on desktop but relied on claude for mobile. No clue for Safari. [19:52] The old Webkit was swapped out for Blink (Chromium's core renderer), which itself started as a fork of Webkit. [19:54] IIRC Safari still uses JavascriptCore rather than V8 for JS code though [19:55] *** scott2 left [19:57] *** scott2 joined [20:19] <[Coke]> anton - any chance yo mught be at the canada raku/perl con '27? [20:42] This is somewhat "too far away" for me to plan. [20:45] "Far away" in time, I see Ontario, Canada to be within a driving distance for me. [20:56] *** kaestralblades joined [21:34] <[Coke]> fair [22:23] *** Sgeo joined [22:49] *** hankache left