| 22 Mar 2023 |
 Wanja Hentze | especially if you want to use this not just for packages but for NixOS modules (any module can do anything, mostly) | 11:43:11 |
 @piegames:matrix.org | * The problem is that we have meta.maintainers which is really fuzzy and also cannot be read without having to evaluate the entire nixpkgs first. IMO there is no way around moving this information outside of the nix code, and to make it per-file/folder instead of per-package | 11:43:12 |
 Alyssa Ross | OfBorg already checks outputs, not files. | 11:43:23 |
| Alyssa Ross | It's annoying, because you have to manually ping people if your changes don't cause rebuilds. | 11:43:33 |
| Wanja Hentze | right, but codeowners style things don't | 11:43:44 |
| Wanja Hentze | so yeah probably something nixpkgs specific is needed | 11:43:59 |
| Alyssa Ross | or gerrit :P | 11:44:06 |
 snowytrees | Ah wasnt sure how ofborg works that makes sense. | 11:44:07 |
| @piegames:matrix.org | In reply to @qyliss:fairydust.space
OfBorg already checks outputs, not files. Oh, is this the reason why it won't ping people if a module with a maintainer is modified? | 11:44:25 |
| Wanja Hentze | ah, gerrit, all the ergonomics of emailing patches with all the openness and flexibility of github | 11:44:52 |
| @piegames:matrix.org | Same for tests IIRC, very annoying | 11:44:54 |
| Alyssa Ross | not sure | 11:44:59 |
| Wanja Hentze | okay that was a bit too harsh on gerrit maybe, but I'm not fond of it | 11:48:47 |
 tea | so what exactly needs to be written? | 12:21:17 |
| tea | given a commit, build the package in the title, check codeowners and then merge it via an action? | 12:22:10 |
| Wanja Hentze | ideally check for review by codeowner *and* CI | 12:23:09 |
| Wanja Hentze | but I think regular CI can already block merging on github | 12:23:41 |
 davidak | In reply to @qyliss:fairydust.space
davidak: then please don't do that again without asking the participants in this specific case i considered it public data because it's publicly available on youtube. they also create a transcript for subtitles and google probably also trains deep learning models with the data and might even sell it (to OpenAI). ChatGPT is of course no solution when the goal is to keep it private. i see 2 options for offline models, but they might take longer to reach the quality
do you think i should have asked before sending the data to OpenAI in this specific case? i will not do that again
| 13:16:21 |
| Alyssa Ross | To me, there's a distinction between something being publicly available (where it could be scraped or whatever), and it being actively sent somewhere. | 13:17:49 |
| Alyssa Ross | it's something that a lot of people seem to feel strongly about, so I think it's safer to ask | 13:19:49 |
| davidak | i see. thank you | 13:22:52 |
|  jurraca joined the room. | 14:05:29 |
 raitobezarius | In reply to @davidak:matrix.org
in this specific case i considered it public data because it's publicly available on youtube. they also create a transcript for subtitles and google probably also trains deep learning models with the data and might even sell it (to OpenAI). ChatGPT is of course no solution when the goal is to keep it private. i see 2 options for offline models, but they might take longer to reach the quality
do you think i should have asked before sending the data to OpenAI in this specific case? i will not do that again
alpaca.cpp / llama.cpp exist with an offline model, of course, it does not have GPT-4 performance | 14:37:29 |
| davidak | yes, i will try them as well | 15:01:29 |
