Time Nick Message 00:46 MTDiscord Why is everyone a bot 00:48 MTDiscord Chat bridge 00:48 Mantar I wasn't a bot the last time I checked 00:48 MTDiscord uh yes you are 00:49 Mantar aw man, when did that happen? 00:49 Mantar stupid robot overlords, uploading my brain without asking first 16:22 MrOvkill Hey! Is anybody opposed to an AI-mod run by an artificial dungeon master, or has it been overdone already? 16:30 ireallyhateirc that entirely depends on what your AI model was trained on 16:31 ireallyhateirc the problem with some stable diffusion models is that they were trained on copyrighted assets without author consent 16:33 ireallyhateirc which is ethically questionable and probably in gray zone in terms of copyright law 16:39 MrOvkill Ah, yes. I wasn't even thinking of image gen, I was more aiming for a dungeon master managing some friendly and some hostile living blocks. Basically, friendly blocks build towns and sell you stuff, etc... All language models. I was going to use Google's Gemma 2 9b on my hardware, provide setup instructions for others to do actually cool stuff once 16:39 MrOvkill it works. 16:46 ireallyhateirc Google mentions what kind of texts the model was trained on but nothing detailed like author info, etc. Well, whatever. As long as the model is freely licensed then it's probably fine. 16:46 ireallyhateirc and people most likely accepted terms of service and license in various proprietary platform and programs that essentially gave Google a green light 16:57 jonadab The case law related to AI training data is really still being worked out. Google is partly counting on the fact that they can afford as many lawyers as anybody. For most people, it is not currently advisable (legally speaking) to publicly distribute anything based on generative-AI output, IMO. IANAL,ATINLA. 17:06 MrOvkill Well, I can tell you as a data scientist myself ( not for Google ) that Google's TOS basically say, in a few million words, "You use our shit and do not have COPPA/PII requirements + an agreement with Google... Yeah they own everything that lives on their servers. There are exceptions, but... Based on Gemini/Gemma's "accidental opinions", it seems 17:06 MrOvkill they're quite keen on being careful "enough". 17:21 jonadab Sure, Google's lawyers can handle the situation however it ends up panning out. 17:22 jonadab I'm not worried that Google as a company will end up being destroyed by some court ruling in an AI-related copyright case. 17:27 MTDiscord MrOvkill: a mod that uses an LLM would typically not contain the llm itself, so its totally fine from a copyright and CDB standpoint. You would probably use the openAI api standard, since many llms use that, both local and remote. 17:27 Mantar I know people think you can guarantee victory with a lot of expensive lawyers, but then two weeks ago Google had a judge declare them an illegal monopoly 17:27 jonadab If you're just calling the API, you should be fine, yeah. 17:28 jonadab Mantar: Watch and see what comes of that. 17:28 jonadab Microsoft has been declared an illegal monopoly half a dozen times over the decades. 17:28 jonadab It does have limits, of course. 17:29 MTDiscord necessarily, such a mod would have a complex installation tag on CDB, and require adding the mod to secure.http_mods 17:29 MTDiscord or whatever the setting is 17:29 MTDiscord has it been done before: not yet, AI is so new 17:29 MTDiscord *llms 17:30 MTDiscord is anybody opposed to it: probably, who cares 17:30 MTDiscord is it allowed on CDB: afaik yes 17:30 jonadab Distributing the code that _implements_ an LLM should also be fine. It's the training data, and the output, that are potentially problematic. 17:31 jonadab If the output are fetched via an API at run time, you're not distributing them. 17:31 jonadab And if the AI is actually remote and you're just calling it via an API, you're not distributing the training data either. 17:32 jonadab So that should all be fine. IANAL,ATINLA. 17:36 MTDiscord jonadab: the current CDB guidance for AI-generated media and code hosted directly on CDB is: you have to declare AI use in your mod (Readme or license) and which models were used. If case law determines that certain AIs, or even all AIs, violate copyright, or something like that, CDB may take your mod down in the future. For now, it's allowed (if declared), and I currently use it for my projects because AI code completion saves about 17:36 MTDiscord 3-4 times the work of debugging and typing. 17:37 MTDiscord some of my projects, especially new ones 21:17 MinetestBot 02[git] 04zmv7 -> 03minetest/minetest: Fix CSM help form using "/" instead of "." (#15034) 13274c223 https://github.com/minetest/minetest/commit/274c223d00519c731fea7de960aac9879273ad8a (152024-08-22T21:15:55Z) 23:53 wsor4035 Sokomine: i know you have effectively abandoned some of your mods, but could you at least cherry pick https://github.com/Sokomine/cottages/pull/19/commits/19df7a72031eae0c4b2f8cb3688a9dbbee33ff54 into cottages when you have a chance so it doesnt crash in 5.9, or review+merge the whole pr? thanks