@lain@crunklord420@BlinkRape did the fucking nasdaq take a thousand point dive today because a chinese company revealed they can make an LLM ai scam cheaper than the USA can do it
@s8n@crunklord420@BlinkRape that's the narrative everyone's going with but i don't believe it. R1 came out over a week ago but investors just realized it monday morning? Neither OpenAI nor Anthropic are public companies, so they are not included in the drop anyway. And there's no reason to believe that cheaper training should lead to LESS demand in GPUs, as now there's a massive number of smaller companies that can afford building a useful LLM, see Jevon's paradox.
@lain@crunklord420@BlinkRape yeah it would make a lot more sense for that to be the result of everyone freaking out about trump saying he was going to put tariffs on semiconductors from taiwan
@s8n@crunklord420@BlinkRape Meta is the competitor with the best open source LLM after deepseek, llama. Here's their stock over the last week, doesn't seem like a big drop to me.
@s8n@crunklord420@BlinkRape it's still news for lots of people as most people aren't paying attention. There's still people here every day talking about how LLMs are useless and stupid parrots and so on.
@lain@crunklord420@BlinkRape yeah we've all been running custom llama finetunes for what a year now and the fact that a powerful desktop computer can do impressive things with an LLM is hardly news
@lain@crunklord420@BlinkRape well to be fair just because they do what they do very impressively does not mean what they do is useful. So far I have yet to see much application for the technology. The text interpreter in stable diffusion forge is extremely impressive though and I think that's an indication of where we're going
@s8n@crunklord420@BlinkRape I can do like a month of my programming work in under a week with Cursor and Claude, so I just can't take it seriously if someone tells me that it's completely useless.
@lain@BlinkRape@s8n if I handed any LLM my game engine and asked it anything related to any actual problems I have it would just hallucinate some javascript.
@lain@crunklord420@BlinkRape I still can't imagine it does anything worthwhile, I've never in my life done any coding where the majority of the work is just writing boilerplate
@lain@crunklord420@BlinkRape I will keep searching for applications and making awesome images though. Icedquinn had taught me a lot about the inner workings and the papers to read to understand things like attention and spike networks with his rants at least
@lain@BlinkRape@s8n I do use it, to do small, usually math related functions and sometimes it's better than google searching. Of course, I spend a lot of time correcting anything it generates.
I'm sure it's useful when your writing code based around a popular API or library, but that's not the type of coding I do.
@lain@BlinkRape@s8n like the random HSV gradient generation function came from ChatGPT. But it also sometimes generates all-white palettes, I never bothered debugging it.
The "programming" it seems capable of doing is copy-pasting popular webshit from Stack Overflow. That's a big industry, and lots of people do it as a full-time job. But it's always just another CRUD app that's no different from the last or the next when you get down to it.
The true zen of programming is making your code such that you don't have boilerplate anymore. The closer you can get to that, the more productive you'll be in reality.
@gentoobro@lain@BlinkRape@s8n >oh you have a question about software rasterization and triangle clipping? here's some 20 year old boilerplate on how to use OpenGL. It's not like I didn't try to get AI to help me. Sometimes it knew stuff, like it knew about old VGA stuff. DOS-era coding which can be hard to find info on these days.
Kinda funny you bring that up, as I had to write a conservative line rasterization algorithm today for lines of arbitrary width and grids of arbitrary size. The internet didn't have any articles on this, at least on the first couple pages of DDG. There were similar but wrong articles, and the details are important.