[00:13:59] https://cdn.discordapp.com/attachments/407537962553966603/1393022474668474510/image.png?ex=6871a8c7&is=68705747&hm=870291bae97db3a7dd84e962ff45f27fd042187dce45bd150f41d41195b9bc67& [00:14:02] > CreateWiki AI [00:14:06] an AI response [00:14:07] ouch [00:14:12] yeah... [00:14:21] powered by ✨ openai ✨ [00:14:39] would be ideal if we didn't need to use an llm to process these, but... [00:14:46] we have too many wiki requests and too few volunteers [00:14:56] yeah [00:15:02] its already bad with the AI carrying [00:15:02] ah [00:15:15] I've edited the request to see if it goes through this time [00:15:20] We had to be pragmatic [00:15:23] AI only touches it once [00:15:30] Anything beyond inital is human review [00:18:19] alright [01:00:48] [1/2] I wonder how much work it would be to support additional backends in CreateWiki and test them for vulnerabilities. OpenAI models are known to be pretty expensive, and they are not even the SOTA right now after Gemini 2.5 pro came out and dominated every benchmark. I just checked the rate limit for its free tier and 5 requests per minute + 100 requests [01:00:49] [2/2] per day should be higher than what we see in the WikiRequestQueue for now (60 requests per day). [01:01:33] If we have to rely on an LLM, I'd prefer we depend on Google's (possibly ill-intended) generosity than paying OpenAI $30/month (or more?). [01:02:15] could always train our own but that's probably not feasible [01:05:04] Yeah even running a local model (e.g. qwen) would be costly with the graphics card investment. I did hear that running them on server CPUs is feasible since someone did get it to work on a 64-core Threadripper (though responses are slow). We already have 40-core CPUs with lots of RAM, so we could potentially leverage that. [01:07:07] The money paid to OpenAI is a small fraction of MH's budget (less than 10%?). It's just frustrating to pay OpenAI with donation money. [01:07:17] does anyone happen to have a bunch of unwanted GPUs or money laying around? :p [01:18:57] [1/2] I think at our scale making an in-house LLM makes little sense with the current number of requests (unless we have some ambitious plan such as offering machine translation services). We should be able to get away with using free-tier apis, and if Google decides to decrease their free tier limit, which they haven't done in the past months, we can just rotate between multiple ba [01:18:57] [2/2] ckends. [01:19:50] The main problem is that we are still handing users' createwiki requests to companies operating LLMs, but at least we're not paying them... [01:24:22] I agree, might be worth considering [02:19:27] We actually will be getting some gpus in a month or two [02:21:42] Both for tinkering with training LLMs/ML models/??? and something with complex/video processing needs [02:23:24] CA or agent can better speak to exact plans [02:27:28] I'll be off my hell week at work in the next few days and will be blasting through queued requests [02:27:50] Hold out hope, folks [02:27:54] nice [02:28:03] I need to get some rest right about now though [03:17:31] Also without the AI responses, you would probably be looking at, at least, a month, if not more for any response to a wiki request, at this point. [03:41:16] I think the request growth curve has gotten less angry, but it is still unforgiving [03:56:50] [1/2] The spike in early 2025 was probably the last straw. [03:56:50] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393078555411480616/image.png?ex=6871dd01&is=68708b81&hm=ee35d2ff5fb9eaedf829c5da68a08668b1cf175603c8390ef0207f297ce47c3b& [04:31:42] [1/2] okay, just imported the navbox template from the miraheze dev wiki and uh [04:31:43] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393087332566958141/Screen_Shot_2025-07-11_at_12.30.54_AM.png?ex=6871e52e&is=687093ae&hm=190032a5c841899c56dbccfab49d9ab6101dc0afc0fedfcea9bf2d8b55fb7f81& [04:31:54] i don't think it's supposed to look like this [04:39:48] oh, wait, i need the navbar too [04:39:55] wait why is that a seperate template [04:40:22] might be because that navbar has other uses than a navbox template [04:44:02] [1/2] okay, nope, still didn't work [04:44:02] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393090433873088692/Screen_Shot_2025-07-11_at_12.43.45_AM.png?ex=6871e811&is=68709691&hm=af33b330e3089b965ba4e1e003c21821f0b2671eb962bd7416df6394940f8257& [04:44:40] [1/2] what's a guy gotta do to get something like this [04:44:40] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393090592841400371/Screen_Shot_2025-07-11_at_12.44.22_AM.png?ex=6871e837&is=687096b7&hm=02d74cf2dd3aef519c93b8b57c1805731d3efe627dd5f0bb798984c60697ab2c& [04:47:01] [1/2] or this [04:47:01] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393091186742263930/Screen_Shot_2025-07-11_at_12.46.38_AM.png?ex=6871e8c5&is=68709745&hm=85c7d569e106f6abdd9d5a094f5d55a3e11daa69815b7192b3b4d0be76454386& [04:49:58] Have you checked those wikis specific templates? [04:50:09] [1/2] HAH, I DID IT [04:50:09] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393091973363138570/Screen_Shot_2025-07-11_at_12.49.51_AM.png?ex=6871e980&is=68709800&hm=1a33926f4e02463a92b40155e55661226fc0dbdfe30591735a5fd9ead9e951e4& [04:50:19] Importing from wikipedia is h a r d [04:50:32] Just had to copy the Navbox module from miraheze meta [04:50:38] > [11/07/2025 14:31] [1/2] okay, just imported the navbox template from the miraheze dev wiki and uh [04:50:46] amelialoveheart: i... don't think that's wikipedia [04:55:34] I’m barely awake rn [05:12:57] fair [05:45:05] html table navbox looks lit [05:58:05] Lots of CSS, though this template doesn't work well on mobile (lots of empty space) [07:05:02] [1/2] How do i stop my template from breaking down like this [07:05:03] [2/2] https://cdn.discordapp.com/attachments/407537962553966603/1393125920646234132/Screenshot_20250711-055454_Chrome.jpg?ex=6872091e&is=6870b79e&hm=8ef51e2fe33a4b0ffb3321dc2f093451a89aa48281352e33766eb9d7635a5bc4& [07:13:32] [1/2] are you looking at Template: page? this is normal behaviour because you aren't using the template, it doesn't have dtaa to show [07:13:32] [2/2] once you use infobox on normal page and provide data for its parameters, it will look like it should [07:18:42] Ahh [07:18:42] Now i understand [07:24:37] What do you mean provide data for its parameters? [07:54:33] [1/5] ``` [07:54:33] [2/5] [07:54:34] [3/5]